System and method for advanced prediction using machine learning and statistical models

Abstract

Systems and methods for implementing advanced statistical models and machine learning algorithms to generate predictions are disclosed. The method includes: receiving a plurality of data from one or more sources; processing the plurality of data to select one or more correlation variables; training one or more predictive models based on the one or more correlation variables and a combination of an advanced statistical model and a machine learning model; evaluating performance of the one or more trained predictive models based on one or more verification techniques; and deploying at least one prediction model to generate a prediction based on the performance.

Description

[0001] Cross-references to related applications

[0002] This application claims priority to U.S. Provisional Application No. 63 / 508,247, filed June 14, 2023, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This disclosure generally relates to the field of machine learning, and more specifically to predictive analytics and forecasting using advanced algorithms and models. Background Technology

[0004] Traditional forecasting models typically rely on historical data and assume relatively stable market conditions, making them ill-equipped to handle sudden and unpredictable changes in consumer behavior. These models lack the flexibility to integrate data—e.g., emerging consumer preferences, market sentiment—which is crucial for adapting to rapid changes, thus hindering their ability to respond to evolving market conditions. Traditional forecasting models depend on numerous assumptions about the data, few of which hold true in real-world data. The static nature of traditional models also means they cannot dynamically adjust to rapid market changes or disruptions, leading to forecasting errors and inventory imbalances. Traditional forecasting methods often struggle to accurately identify and handle outliers in the data, resulting in forecasting bias and inefficient inventory management. Such traditional forecasting methods may face scalability challenges when dealing with large datasets or complex product categories, leading to increased processing time and decreased forecasting accuracy. Advanced forecasting methods are needed that can dynamically adapt to changing market conditions, integrate data sources, and effectively capture the complexity of modern consumer behavior. Summary of the Invention

[0005] According to various aspects of this disclosure, systems and computer implementations for adaptive forecasting using advanced statistical models and machine learning algorithms are disclosed.

[0006] In some embodiments, the computer-implemented method includes: receiving multiple data from one or more sources by one or more processors; processing the multiple data by one or more processors to select one or more relevant variables; training one or more predictive models by one or more processors based on one or more relevant variables and a combination of advanced statistical models and machine learning models; evaluating the performance of one or more trained predictive models by one or more processors based on one or more validation techniques; and deploying at least one predictive model based on performance by one or more processors to generate one or more predictions.

[0007] In some embodiments, a system includes one or more processors of a computing system; and at least one non-transitory computer-readable medium storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising: receiving a plurality of data from one or more sources; processing the plurality of data to select one or more relevant variables; training one or more predictive models based on the one or more relevant variables and a combination of advanced statistical models and machine learning models; evaluating performance of the one or more trained predictive models based on one or more validation techniques; and deploying at least one predictive model based on the performance to generate one or more predictions.

[0008] In some embodiments, a non-transitory computer-readable medium stores instructions that, when executed by one or more processors of a computing system, cause the one or more processors to perform operations comprising: receiving a plurality of data from one or more sources; processing the plurality of data to select one or more relevant variables; training one or more predictive models based on the one or more relevant variables and a combination of advanced statistical models and machine learning models; evaluating performance of the one or more trained predictive models based on one or more validation techniques; and deploying at least one predictive model based on the performance to generate one or more predictions.

[0009] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the detailed embodiments claimed.

[0010] BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate various example embodiments and together with the description, explain the principles of the disclosed embodiments.

[0012] Figure 1 The ability to implement advanced statistical models and machine learning algorithms to generate predictions in accordance with various aspects of the present disclosure is introduced.

[0013] Figure 2A is a flowchart illustrating a machine learning enhanced forecasting process in accordance with various aspects of the present disclosure.

[0014] Figure 2B is a chart illustrating the data preparation step of a machine learning enhanced forecasting process in accordance with various aspects of the present disclosure.

[0015] Figure 2C is a chart illustrating the necessity of the exogenous phase of a machine learning enhanced forecasting process in accordance with various aspects of the present disclosure.

[0016] Figure 2Dis a chart illustrating one or more criteria for selection of model inputs for the prediction exogenous phase of the machine learning enhanced prediction process, in accordance with various aspects of the present disclosure.

[0017] Figure 2E is a chart illustrating the Homados component of the machine learning enhanced prediction process, in accordance with various aspects of the present disclosure.

[0018] Figure 2F is a chart illustrating the training model step of the machine learning enhanced prediction process, in accordance with various aspects of the present disclosure.

[0019] Figure 2G is a chart illustrating the training model phase of the process for capturing and predicting time series data, in accordance with various aspects of the present disclosure.

[0020] Figure 2H is a chart illustrating the training model phase of the process in which an LSTM network is utilized, in accordance with various aspects of the present disclosure.

[0021] Figure 2I is shown a training model phase of the process utilizing a plurality of models to ensure robust and accurate predictions, in accordance with various aspects of the present disclosure.

[0022] Figure 2J is a chart illustrating the production phase of the machine learning enhanced prediction process, in accordance with various aspects of the present disclosure.

[0023] Figures 2K to 2R is a chart illustrating the prediction phase of the machine learning enhanced prediction process, in accordance with various aspects of the present disclosure.

[0024] Figure 2S is a chart illustrating the prediction phase of the machine learning enhanced prediction process, in accordance with various aspects of the present disclosure.

[0025] Figures 2T to 2W is a chart illustrating a comparison of prediction accuracy of the prediction model to actual sales data and present conditions, in accordance with various aspects of the present disclosure.

[0026] Figure 3 is a flowchart of a process for adaptive prediction using advanced statistical models and machine learning models, in accordance with various aspects of the present disclosure.

[0027] Figure 4 is shown an example machine learning training flowchart.

[0028] Figure 5 is shown an implementation of a computer system that performs the techniques described herein. DETAILED DESCRIPTION

[0029] While the principles of the present disclosure are described herein with reference to illustrative embodiments in a particular application, it is understood that the present disclosure is not limited thereto. Those having ordinary skill in the art having the benefit of the teachings presented in the specification are expected to recognize alternatives, applications, embodiments, and equivalents fall within the scope of the embodiments described herein. Accordingly, the present disclosure should not be considered as limited by the foregoing description.

[0030] Various non-limiting embodiments of the present disclosure will now be described in order to provide a thorough understanding of the principles of the advanced forecasting system, its structure, functionality, and use. These embodiments can encompass the integration of machine learning algorithms (e.g., Long Short Term Memory (LSTM) and Random Forest) with statistical methods (e.g., ARIMA and Exponential Smoothing) to predict future values and accuracy. Furthermore, these embodiments detail the implementation of automated data pipelines, on-demand forecasting capabilities, scalable architecture, anomaly detection, and exogenous variable forecasting.

[0031] Conventional forecasting models often fail to account for the inherent variability in demand, resulting in oversimplified forecasts that do not adequately capture fluctuations caused by factors such as economic instability. Conventional forecasting models are technically difficult to incorporate periodic updates, resulting in forecasts that quickly become outdated as new data emerges. Conventional forecasting models often underperform in extreme market conditions (e.g., a pandemic or sudden surge in demand) due to their reliance on stable historical patterns. Conventional forecasting methods often lack the ability to learn from new data inputs in an automated manner, requiring manual recalibration to adapt to new trends and patterns.

[0032] Conventional forecasting methods typically rely on predefined structures that can not be flexible enough to adapt to changing data patterns, limiting their ability to improve over time. Conventional forecasting methods are technically difficult to incorporate categorical variables (e.g., product categories, customer segments), resulting in oversimplified forecasts that do not capture the full complexity of the patterns. Such conventional forecasting models are sensitive to noise and anomalies in the data, which can bias the results and lead to inaccuracies.

[0033] To address the limitations of traditional forecasting models, the system 100 can leverage advanced machine learning algorithms to capture complex non-linear relationships in the data, improving the accuracy of the forecasting model. The system 100 can integrate diverse datasets (e.g., economic indicators, market sentiment, or social media trends) for accurate predictions and dynamically adjust as new data becomes available. The system 100 can efficiently handle large-scale datasets using distributed computing and parallel processing, ensuring scalability. The system 100 can automatically learn from new data and adapt, continuously improving its prediction accuracy. By capturing periodic trends and underlying variables, the system 100 can provide a granular dynamic view of demand for each product, enabling precise allocation of production resources to effectively reduce excess and alleviate shortages. The system 100 can integrate time-series modeling and machine learning models to capture inherent complex temporal patterns and interdependencies in the data. By combining historical data with data streams, the forecasting model can generate forecasts that dynamically adapt to changing market dynamics. This integrated approach not only improves the accuracy of predictions but also facilitates agile decision-making in resource allocation, effectively optimizing production to minimize excess and alleviate shortages.

[0034] Figure 1 The ability to implement advanced statistical models and machine learning algorithms to generate predictions in accordance with various aspects of the present disclosure is introduced. Figure 1 An example architecture (i.e., one example embodiment of one or more example embodiments of the present disclosure) includes a system 100 that includes an adaptive forecasting system 101 and a data source 103.

[0035] In one embodiment, the adaptive forecasting system 101 can be a platform with multiple interconnected components. The adaptive forecasting system 101 can include one or more servers, intelligent network devices, computing devices, components, and corresponding software for generating predictions using advanced statistical models and machine learning algorithms. Furthermore, it is noted that the adaptive forecasting system 101 can be a standalone entity in the system 100.

[0036] The adaptive forecasting system 101 can aggregate and clean data from various sources, handle missing values, normalize data, and perform feature engineering in order to prepare high-quality inputs for the prediction models (i.e., forecasting models). The adaptive forecasting system 101 can use statistical and machine learning techniques to identify and select the most relevant features, ensuring that only significant variables are incorporated into the prediction models to improve accuracy. The adaptive forecasting system 101 can train multiple prediction models, such as ARIMA, LSTM, and tree-based models, and optimize their hyperparameters using techniques such as Bayesian optimization to achieve optimal performance. The adaptive forecasting system 101 can rigorously validate using cross-validation and backtesting, and can evaluate model performance using various metrics to ensure robustness and prevent overfitting. The adaptive forecasting system 101 can deploy the best-performing prediction models into production, integrate them with business systems, and continuously monitor their performance, making necessary adjustments to maintain accuracy and reliability. For example, the adaptive forecasting system 101 can implement on-demand forecasting capabilities to generate up-to-date forecasts based on the latest data and continuously update the prediction models as new data becomes available to maintain accuracy and relevance under dynamic market conditions. The adaptive forecasting system 101 can leverage scalable architecture and parallel processing techniques to handle large volumes of data and speed up the training and evaluation of prediction models. The adaptive forecasting system 101 can incorporate exogenous variables (such as economic indicators and market trends) into the forecasting to enhance model inputs and improve the accuracy of forecasts. Additionally or alternatively, the adaptive forecasting system 101 can integrate anomaly detection mechanisms to identify and handle outliers or unusual patterns in the data.

[0037] In one example, the data sources 103 can include various internal and external data sources that can provide comprehensive insights into past and future trends. In one example, internal data sources can include historical sales data, inventory levels, and customer transactions, which can be stored in databases (e.g., customer relationship management (CRM) databases) or enterprise resource planning (ERP) systems. In one example, external data sources can include market trends, economic indicators, social media sentiment, and weather data, which can be accessed through public databases, third-party providers, and APIs. In one example, these diverse data inputs can be integrated into a centralized database or data warehouse to ensure that all relevant information is available for analysis. By integrating these data sources into a unified system, the adaptive forecasting system 101 can leverage advanced analytics and machine learning techniques to generate accurate and actionable forecasts.

[0038] In one example, the adaptive forecasting system 101 can include a data preparation module 105, a feature selection module 107, a model selection and training module 109, a validation and evaluation module 111, a forecasting module 113, an integration and deployment module 115, a visualization module 117, a monitoring and maintenance module 119, or any combination thereof. As used herein, terms such as “component” or “module” generally refer to hardware and / or software, e.g., a processor, etc., for implementing the associated functionality. It is contemplated that the functionality of these components is combined in one or more components, or performed by other components with equivalent functionality.

[0039] In one example, the data preparation module 105 can collect relevant data from various data sources (e.g., data sources 103) through various data collection techniques. In one example, the data preparation module 105 can use a web crawler component to access the data sources 103 to collect relevant data. In one example, the data preparation module 105 can include various software applications (e.g., eXtensible Markup Language (XML) data mining applications) that automatically search and return relevant data. Once collection is complete, the data can undergo a rigorous pre-processing phase to ensure that it is clean, consistent, and ready for analysis. This can involve handling missing values (which can be filled using various imputation methods) and resolving inconsistencies or errors in the data. In addition, the pre-processing phase can include feature engineering (where new features can be created to enhance the predictive power of the model) and data transformation (such as normalization and smoothing) to ensure that the data format is suitable for model training. By thoroughly preparing the data, the data preparation module 105 lays a solid foundation for accurate and reliable forecasting.

[0040] In one example, the feature selection module 107 can identify relevant and influential variables from the dataset that should be included in the predictive model. The feature selection module 107 can perform a comprehensive analysis of the dataset, taking into account historical features and exogenous features such as sales, inventory levels, economic indicators, and weather data. Advanced statistical methods and machine learning techniques (such as correlation analysis, mutual information, and recursive feature elimination) can be employed to assess the importance of each feature. The feature selection module 107 can retain features that can have a significant contribution to the predictive accuracy of the model, while discarding those that add noise or redundancy. This can improve the performance of the predictive model and can reduce computational complexity.

[0041] In one example, the model selection and training module 109 can identify suitable predictive models and fine-tune them for optimal performance. The model selection and training module 109 can evaluate various forecasting models, such as ARIMA, Seasonal AutoRegressive Integrated Moving Average with exogenous variables (SARIMAX), Facebook Prophet, LSTM, and tree-based models (such as random forest or gradient boosting), to determine which models are best suited for the characteristics of the data and the forecasting objective. Once a predictive model is selected, the training phase can include inputting pre-processed data into these models and adjusting their parameters to improve accuracy. This process can include hyperparameter tuning, which can be automated and parallelized using techniques such as Bayesian optimization and tools such as Hyperopt. In one example, in an LSTM model, hyperparameters such as the number of layers, the number of units per layer, and the learning rate can be tuned. In one example, in a tree-based model, parameters such as the depth of the trees, the number of trees, and the minimum number of samples per leaf can be adjusted. The training phase is iterative, involving continuous evaluation and improvement based on a validation dataset to ensure that the model generalizes to unseen data. In one example, advanced machine learning techniques such as cross-validation, ensemble methods, and neural networks can be utilized to build robust models that can capture complex patterns and relationships in the data.

[0042] In one example, the validation and evaluation module 111 can implement machine learning techniques to ensure the reliability and accuracy of the predictive models. The validation and evaluation module 111 can perform rigorous evaluation of the trained models using an independent validation dataset that is not part of the training process, providing an unbiased assessment of the model performance. The validation and evaluation module 111 can utilize advanced machine learning techniques to perform cross-validation, where the data is split into multiple folds, and the model is trained and validated on different subsets to ensure robustness and prevent overfitting. The evaluation process can involve backtesting, where historical data can be used to simulate the forecasting performance in real-world scenarios. By thoroughly validating the models, the validation and evaluation module 111 can identify any biases and can facilitate further improvement and tuning. Such validation and evaluation processes can ensure that the models are highly accurate and also generalize to new data.

[0043] In one example, the forecasting module 113 can generate accurate predictions of future demand based on historical data and other relevant factors. The forecasting module 113 can utilize advanced statistical methods and machine learning algorithms to analyze patterns and trends in the data and extrapolate them into the future. In one example, time series forecasting techniques such as ARIMA, SARIMA, and exponential smoothing can be commonly used to model temporal dependencies in the data and make short-term predictions. For long-term forecasting and scenarios involving complex relationships, machine learning models such as LSTM neural networks, gradient boosting machines, and random forests can be employed. These models can capture non-linear relationships and interactions among multiple variables, resulting in accurate forecasts. In one example, the forecasting module 113 can also consider external factors such as market trends, economic indicators, and seasonal patterns to improve the accuracy of the predictions. In one example, by continuously monitoring model performance and incorporating feedback from actual sales data, the module ensures that the predictions remain up-to-date and reliable.

[0044] In one example, the integration and deployment module 115 can facilitate the integration of the forecasting models and associated data pipelines into existing business systems such as enterprise resource planning (ERP), supply chain management (SCM), CRM, and others. By integrating with these systems, the forecasting models can leverage relevant data sources and provide insights directly to decision-makers. The integration and deployment module 115 can manage the deployment of the forecasting models into production environments, ensuring scalability, reliability, and performance. This can involve setting up automated workflows for data ingestion, preprocessing, model training, validation, and prediction, as well as implementing monitoring and alerting systems to detect and address any issues.

[0045] In one example, the visualization module 117 can provide visual presentations of the results and trends of the forecasts. The visualization module 117 can utilize various visualization techniques, including charts, graphs, dashboards, and heatmaps, to present the forecasted data in a clear and meaningful manner. In one example, time series plots can illustrate historical sales trends and forecasted values over time, while scatter plots can reveal correlations between different variables. In one example, interactive dashboards can allow users to explore the forecasted results from different perspectives and drill down into specific regions or product categories. The visualization module 117 can enable comparisons of multiple forecasting scenarios, aiding scenario planning and risk management. By providing actionable insights in a visually appealing format, the module can enhance the usability and effectiveness of the forecasting system.

[0046] In one example, the monitoring and maintenance module 119 can continuously monitor the accuracy and effectiveness of the deployed model in generating forecasts by comparing the predicted values with actual outcomes. Any deviations or discrepancies are promptly identified and investigated, allowing for timely adjustments and improvements to the model. In one example, the monitoring and maintenance module 119 can track various performance metrics to measure the overall effectiveness of the forecasting model. In one example, periodic maintenance tasks (e.g., model retraining and updates) can be performed to ensure that the model remains relevant and adapts to changing market conditions. Furthermore, the monitoring and maintenance module 119 can incorporate a feedback loop from stakeholders and end-users to gather insights and suggestions for improving the forecasting process.

[0047] The above-described modules and components of the adaptive forecasting system 101 can be implemented in hardware, firmware, software, or a combination thereof. The various implementations presented herein contemplate any and all arrangements and models.

[0048] Figures 2A to 2S is a diagram illustrating a machine learning-enhanced forecasting process in accordance with various aspects of the present disclosure. In various embodiments, the adaptive forecasting system 101 and / or any of the modules 103-111 can perform one or more processes of Figures 2A to 2S and are implemented using a chip set including a processor and a memory as illustrated in Figure 5 Thus, the adaptive forecasting system 101 and / or any of the modules 103-111 provide means for implementing various portions of Figures 2A to 2S and means for implementing embodiments of other processes described herein in conjunction with other components of the system 100.

[0049] Figure 2A is a flowchart illustrating a forecasting workflow 200 using machine learning in accordance with various aspects of the present disclosure. In one example, the forecasting workflow 200 can be an end-to-end process that can receive a base time series data frame and can generate an accurate model.

[0050] In block 201, several key steps are taken in preparing the data for analysis to ensure its quality and usability. Initially, missing dates are addressed by appending them to the dataset, ensuring full temporal coverage. Missing values are then filled using appropriate techniques (e.g., imputation or interpolation) to maintain data integrity. Smoothing methods (e.g., moving average or exponential smoothing) can be applied to reduce noise and reveal underlying trends in the data. Additionally, computed labels (e.g., seasonal indicators or trend components) are incorporated to provide further insights into potential patterns. Finally, the dataset is divided into training and testing sets to facilitate model development and evaluation, ensuring that the prediction algorithm generalizes well to unseen data.

[0051] In block 203, in the pre-estimation task, exogenous features play a crucial role in enhancing the predictive power of time series models. These features are typically estimated outside the main dataset using appropriate methods such as machine learning algorithms or statistical models. Subsequently, a performance threshold is established to evaluate the quality of these estimates, ensuring that they meet predefined accuracy and reliability standards. Features whose estimates exceed this threshold are considered suitable candidates and are incorporated as inputs to the time series model. By incorporating these exogenous factors, which can capture external influences such as economic indicators or market trends, time series models gain additional explanatory power and can generate more robust and accurate predictions.

[0052] In block 205, in the prediction modeling, particularly in machine learning modeling, feature importance simulations play a crucial role in identifying the most relevant variables to incorporate as model inputs. These simulations involve systematically assessing the impact of each feature on model performance. By iteratively evaluating the model's performance with and without a particular feature, statistical methods determine the relative importance of each variable in explaining the variance of the target variable. Features that consistently demonstrate a significant impact on the model's predictive accuracy are considered statistically significant and are prioritized as inputs to the final model.

[0053] In block 207, multiple time series models are developed by parallelizing the process of model training and hyperparameter tuning. This approach utilizes parallel computing to train various models, such as ARIMA and LSTM networks, each with different hyperparameter configurations simultaneously. By exploring a wide range of model structures and parameter settings, this method identifies the most effective combinations for accurate estimation. Throughout the process, all models, their corresponding hyperparameters, and performance metrics are meticulously documented using ML flow, an open-source platform for managing the machine learning lifecycle. ML flow facilitates tracking experimental results, enabling easy comparison and selection of the best-performing models.

[0054] In block 209, once the best-performing model is determined through rigorous evaluation and hyperparameter tuning, it is pushed into production. This deployment phase involves integrating the selected model into a runtime environment where it can process data and generate accurate estimates. These models are continuously monitored to ensure their performance and accuracy remain consistent in a dynamic environment.

[0055] In block 211, the deployed models are loaded and utilized to generate on-demand estimates. These models, which have been rigorously validated and tuned, process incoming data to accurately predict future patterns. The generated estimates are then used to inform decision-making processes, enabling timely adjustments to production planning, inventory management, and resource allocation.

[0056] Figure 2Bis a chart illustrating the data preparation step 213 of the machine learning enhanced forecasting process, in accordance with various aspects of the present disclosure. In one example, the adaptive forecasting system 101 can clean the data through user-selected transformations, providing flexibility to address various data quality issues. The user can select to append missing dates to ensure temporal continuity, impute missing values using their preferred method (e.g., mean, median, or advanced imputation techniques), and apply their selected smoothing method to selected columns to reduce noise and highlight trends. Additionally, the user can choose to skip these transformations if deemed unnecessary. This customizable approach ensures that the data is accurately prepared according to specific needs and preferences.

[0057] In one example, the adaptive forecasting system 101 can generate a raw table containing time series data. The time series data can include a date column, columns that form groups (e.g., customer-SKU pairs), a target variable column, and additional feature columns (e.g., exogenous features). If a customer skips a week between orders or exhibits highly erratic order behavior (e.g., extremely high values followed by extremely low values), such events can result in poor performance of the forecast. To mitigate this, standard time series data pre-processing and cleaning is performed, ensuring that the data structure is sound and ready for accurate analysis and forecasting.

[0058] In one example, chart 215 depicts time series data that can be difficult to model. There are missing values or values that are highly erratic (e.g., high variance), and in a modeling framework, there should not be any missing values. Traditional methods can replace missing values with 0, but the subsequent sequence can appear disordered. Traditional methods can also fill missing values with previous values or an average of previous values, but the subsequent sequence can exhibit sharp spikes. Thus, the adaptive forecasting system 101 can implement various smoothing techniques (e.g., replace values with an average of the previous 7 or 14 days of data) to make the sequence more interpretable and modelable (e.g., chart 217).

[0059] Figure 2C is a chart illustrating the forecasting exogenous features stage 219 of the machine learning enhanced forecasting process, in accordance with various aspects of the present disclosure. In the forecasting exogenous features stage 219, the focus is on generating predictions of external factors that can influence the forecast. These exogenous features, which can include variables such as economic indicators, weather patterns, or marketing campaigns, play a key role in improving the accuracy and granularity of the predictive model. Using advanced forecasting techniques (e.g., machine learning algorithms or statistical models), these features are independently forecasted to capture their potential impact on future trends.

[0060] In one instance, after the initial data preparation phase, the adaptive prediction system 101 can proceed to identify the optimal model architecture and determine the best model input set. Given that the prepared dataset may contain massive amounts of data, the system employs an automated method to select only relevant and worthwhile exogenous features to be included as input to the predictive model. Through this automated feature selection process, the adaptive prediction system 101 can aim to simplify the modeling process and improve the model's predictive accuracy by focusing only on the most influential variables. During this process, the adaptive prediction system 101 can perform various calculations to determine the optimal architecture and feature set, such as: Formula 221 is y A simple linear model. It has two input variables. x and z Coefficient beta1 ( β 1) and beta2 ( β 2) and constant error term e .

[0061] Formula 223 is the sales amount (i.e., y A simple linear model of ). It has two input variables: inventory and oil price, and coefficient beta1 ( β 1) and beta2 ( β 2) and constant error term e Inventory and oil prices are examples of exogenous characteristics because they are independent of sales volume.

[0062] Dividing constant error term e In addition, the terms in Formula 225 change over time. For example, today's sales are a multiple of today's inventory plus a multiple of today's oil price plus a constant term. e Assuming today is the [number]th [day]... t The formula shows that sales two days ago (today – 2) equals the combination of inventory two days ago and oil price two days ago.

[0063] Formula 227 states that sales one day prior equal the combination of inventory one day prior and oil price one day prior.

[0064] Formula 229 states that today's sales revenue is a combination of today's inventory and today's oil price. The values ​​of sales revenue, inventory, and oil price are up to today (i.e., time t). Machine learning algorithms can use a constant coefficient beta1 (…). β 1) beta2 ( β 2) and error terms e Find the optimal value. To make the sales model practical, the adaptive forecasting system 101 can use it to predict sales at future times.

[0065] To use this model to predict tomorrow's sales, the adaptive forecasting system 101 can substitute the inventory value for tomorrow and the oil value for tomorrow into equation 231.

[0066] To obtain a prediction of sales two days from now, the adaptive forecasting system 101 can substitute the inventory value two days from now and the oil value two days from now into equation 233. To do this, the adaptive forecasting system 101 can have estimates for tomorrow and the day after tomorrow for exogenous variables.

[0067] Through iterative optimization techniques, the adaptive forecasting system 101 can evaluate different combinations of features and model structures to identify the configuration that produces the most accurate and robust predictions. This meticulous computational process ensures that the final model architecture effectively captures the underlying relationships within the data, enabling precise predictions of future patterns.

[0068] Figure 2D is a chart illustrating one or more criteria for selecting model inputs for the exogenous phase of the machine learning-enhanced forecasting process, in accordance with various aspects of the present disclosure. In one example, the adaptive forecasting system 101 can evaluate potential external factors to determine whether they are suitable for inclusion in the prediction model. This evaluation is based on two key criteria. First, the adaptive forecasting system 101 can assess whether an exogenous feature is predictable (235), meaning that it can be accurately estimated using available data and reliable methods. Second, the adaptive forecasting system 101 can examine whether the exogenous feature makes a significant contribution to the performance of the model (237), thereby improving its accuracy and robustness. Only those features that satisfy both criteria (by demonstrating both predictability and a meaningful impact on model performance) are included in the final model. This selective approach ensures that the prediction model is both precise and efficient, leveraging the most relevant external factors to improve predictions.

[0069] Figure 2Eis a chart showing the Homados component 239 of the machine learning enhanced forecasting process according to various aspects of the present disclosure. The adaptive forecasting system 101 can run simulations to obtain feature importance scores for potential model inputs. This process involves creating virtual features using randomly sampled numbers as a benchmark for comparison. The adaptive forecasting system 101 can generate a list of features that contribute more to model performance than random noise. In one example, by running approximately 500 simulations (each involving additional white noise virtual features sampled from different types of distributions), the adaptive forecasting system 101 can determine the feature importance scores for each potential input. Features that exhibit statistically higher importance scores than white noise virtual features are identified as important contributors and are included in the final model. These simulations and feature selection are determined for each unique time series to be modeled. After the simulations, a list is completed consisting of features that contribute more to model performance than random noise.

[0070] In one example, chart 241 shows an exogenous feature that exhibits a distribution statistically equivalent to the highest scoring white noise feature. The two distributions are indistinguishable from one another, indicating that this exogenous feature does not provide any meaningful information or predictive power beyond what random noise can contribute. Thus, including this feature in the model does not add any value and can detract from the performance of the model. In one example, chart 243 is an example of an exogenous feature (first lag of target value (number of units ordered)) that has a feature importance score statistically higher than the highest scoring white noise feature. Thus, this feature can be included in the final model. These examples highlight the importance of rigorous feature selection, ensuring that only those variables with true predictive significance are included in the final forecasting model.

[0071] In one embodiment, the Homados component 239 can evaluate the importance of each predictable exogenous feature by running modeling simulations. The steps can include, but are not limited to: 1. Obtain white noise features by randomly sampling from normal, Poisson, and binomial distributions; 2. Calculate lagged values of target value, predictable exogenous features, and white noise features - these are the input features; 3. Train a random forest model using the input features described above; 4. Extract permutation feature importance scores for all input features; 5. Repeat 500 times, and then... 6. Create a list of all exogenous features that exhibit statistically higher feature importance scores than the average highest scoring white noise feature, as well as the target value lags.

[0072] Figure 2Fis a diagram illustrating the training model step of the machine learning enhanced forecasting process according to various aspects of the present disclosure. In the training model stage 245, three different types of models can be developed to ensure robust and accurate forecasting. The first model type is an autoregressive integrated moving average model (ARIMA) 247, which effectively captures linear time dependencies in time series data. In one example, the ARIMA 247 can be composed of an autoregressive component and a moving average component. Each component models the time series by a linear combination of past values of the target variable or error term. Exogenous variables can also be included in these models (e.g., ARIMAX model). The second model type is a long short-term memory network (LSTM) 249, which is a type of recurrent neural network that is good at learning and predicting complex patterns and long-term dependencies in sequential data. The third model type is a tree-based model 251, such as a random forest or gradient boosting machine, which is very powerful at handling non-linear relationships and interactions among features. By training these diverse models, the process leverages the strengths of each method, thereby improving the accuracy and reliability of the overall forecasting.

[0073] ARIMA 247 is typically used for short-term but not long-term forecasting, as its forecasts can either converge to a stable value (e.g., stabilize at a value over time) or uncontrollably diverge (e.g., become extremely large as future time extends) over time. The reliability of ARIMA 247 forecasts depends on the assumption that the residuals are uncorrelated and normally distributed; if this assumption is violated, the prediction intervals can become unreliable. Furthermore, ARIMA 247 forecasts are prone to trend errors when there is a change in trend near the end of the training period and seasonality is not accounted for. The architecture of this model is defined by three parameters: (i) the order of the autoregressive part (p), (ii) the degree of differencing to make the time series stationary (d), and (iii) the order of the moving average part (q). These parameters are crucial for specifying the structure and behavior of the ARIMA 247 model and determining whether it is suitable for capturing the underlying patterns in the time series data.

[0074] Figure 2Gis a diagram illustrating the training model stage 245 of the process for capturing and forecasting time series data according to various aspects of the present disclosure. Specifically, the ARIMA 247 can include an AR(p) model 253 that uses a specified number of lag observations as inputs to forecast future values. These autoregressive models are good at modeling momentum and persistence in time series data. In addition, the ARIMA 247 can include an MA(q) model 255 that can utilize past forecasted errors in a moving average process to improve predictions. By integrating these components, the ARIMA 247 can balance the influence of past values (AR) with the influence of past prediction errors (MA), thereby improving the accuracy and robustness of short-term forecasts. This dual approach can allow the process to leverage the strengths of both autoregressive and moving average techniques to capture potential patterns and dependencies in the data.

[0075] Figure 2H is a diagram illustrating the training model stage 245 of the process in which an LSTM network 249 is utilized according to various aspects of the present disclosure. Neural network models, such as the LSTM network 249, have been shown to outperform ARIMA models in many scenarios. LSTMs generally perform well when there is a large amount of data, although effective and efficient hyperparameter tuning can be difficult to perform. The LSTM network 249 is a special type of recurrent neural network (RNN) that is widely used for time series forecasting. The LSTM network 249 is particularly effective because it is able to maintain long-term dependencies in data through its unique gating architecture. These gates control the flow of information, allowing the LSTM network 249 to retain, forget, or ignore data points based on a probabilistic model. For example, the hidden state h t , cell state A and input X t may evolve through a sequence 257: h 0 , A, X 0 ；h 1 , A, X 1 ；h 2 , A, X 2 The sequence of past values of a variable (e.g., daily oil prices and inventory levels for the past week, denoted as h t , A , X t) can be inputted into the LSTM network 249 to predict future sales. Using a series of gates (each gate having its own RNN), the LSTM network 249 can decide to keep, forget, or ignore data points in a probabilistic manner, thereby improving its predictions. After each prediction, the output is fed back into the model to predict the next value in the sequence, thereby enabling the LSTM network 249 to learn from past predictions and iteratively improve its prediction accuracy.

[0076] Figure 2I A training model phase of the flow is shown, in accordance with various aspects of the present disclosure, which utilizes a plurality of models to ensure robust and accurate forecasting. In one example, the current model list 259 can include SARIMAX, Facebook Prophet, PyTorch LSTM, quantile regression, and tree-based models. Each of these models can provide unique advantages in handling different aspects of time series data and forecasting challenges.

[0077] In one example, the SARIMAX model can be a generalization of the ARIMA model that takes into account seasonality and exogenous variables. In addition to the parameters (p, d, q), it has seasonal versions of P, D, Q, and a season length parameter m. The SARIMAX model can be used to predict in the short-term forecasting range and can be superior to the ARIMA model in capturing effects due to seasonality, but can still be susceptible to errors from outliers near the end of the training period.

[0078] In one example, the Facebook Prophet model can model time series as a curve-fitting problem. The model can take the form of an additive model, i.e., it is a sum of functions that capture different phenomena in the time series. The Facebook Prophet can have several advantages over ARIMA, such as: (i) it can accommodate seasonality with multiple periods; (ii) it can accommodate new additive modeling components; (iii) it does not require regular intervals of data points and does not need to impute missing values when removing outliers; (iv) it can be quickly fitted using backfitting or the L-BFGS algorithm of Stan; and, (v) it can have interpretable parameters that capture elements such as trends, seasonality, holidays / special events, etc.

[0079] In one example, the quantile regression model can approach probabilistic forecasting, which provides the forecast estimate in the form of a probability distribution. For example, for each time point in the forecasting range t , the 2nd, 10th, 25th, 75th, 90th, and 98th quantiles are estimated. The interpretation of this forecast is different from that of the ARIMA model, in which the forecast is interpreted as falling within a certain range with a certain probability.

[0080] In the supervised learning framework, tree-based models are a common complement to parametric models. The appeal of using tree-based models is the flexibility that the model provides in handling non-linear data. Tree-based models can use a non-parametric approach to fit a line to the data, which can make the model more flexible than its linear and parametric counterparts. In one instance, the adaptive forecasting system 101 can incorporate five types of tree-based modeling algorithms: random forests, gradient boosting, histogram gradient boosting, Extra Trees, and Adaboost. Each of these time series modeling algorithms can handle problems in different and unique ways. The performance of these models can determine which approach to select. In one instance, each time a new time series modeling technique is developed, the adaptive forecasting system 101 can add these new time series modeling techniques.

[0081] In one instance, the adaptive forecasting system 101 can perform hyperparameter tuning 261 using a software package named Hyperopt. Hyperparameter tuning is a key aspect of optimizing model performance, and the system 101 can employ Bayesian optimization for this purpose. Hyperopt is an open-source Bayesian optimizer that tries to minimize the loss as a function of all model hyperparameters in a Bayesian way (i.e., it is not just a random search of a grid of parameter values, but intelligently learns which combinations of values perform well in the run and focuses the search on this, essentially performing an intelligent grid search). In one example, as the model is built, Hyperopt learns its loss and determines which combinations of hyperparameters seem promising and which seem ineffective. It only explores promising combinations deeply, which saves a lot of time. Hyperopt can also integrate with Spark, so that modeling jobs can run in parallel on a Spark cluster, rather than just on one machine. In the process of finding the best model, the adaptive forecasting system 101 can track the results from the Hyperopt experiments. This process uses Hyperopt to achieve automation and parallelization, enabling efficient and exhaustive exploration of the hyperparameter space to identify the best configuration for each model.

[0082] Figure 2JThis is a diagram illustrating the production phase of a machine learning-enhanced prediction workflow according to various aspects of this disclosure. In production phase 263, MLflow can be used to manage and streamline the entire lifecycle of a machine learning model. MLflow can track the model training process and hyperparameter tuning, ensuring that all experiments and their results are meticulously documented. Once the optimal model is determined, it is registered in a central location within MLflow, making it easily accessible and manageable. This registration process enables seamless deployment of the model in a production environment for use. Workflow 265 can include training, tracking, registration, production, and prediction phases, ensuring that the model is not only developed and optimized but also efficiently deployed to provide on-demand predictions.

[0083] In one example, the training phase of Workflow 265 can initialize experiments for each group and model type and train all prediction models using the training set. In another example, the tracking model phase of Workflow 265 can perform hyperparameter tuning within each experiment and record each run. The tracking model can stop running when the loss metric reaches its minimum. In another example, the registration phase of Workflow 265 can filter all model runs, rank models based on out-of-sample performance, and register the best model to the MLFlow model repository. In another example, the production phase of Workflow 265 can deploy the most recently registered model to production and move all previous models to archive. In another example, the prediction phase of Workflow 265 can pull models registered to production and use them to generate predictions. In one instance, MLflow can handle four main functions: 1. Track experiments to record and compare parameters and results; 2. Package ML code in a reusable and reproducible form for sharing with other data scientists or for transfer to production; 3. Manage models from various ML libraries and deploy them to various model services and inference platforms; and 4. Provide a central model repository for collaborative management of the entire lifecycle of MLflow models, including model version control, stage transformations, and annotation.

[0084] Figures 2K to 2R This is a diagram illustrating the prediction phase of a machine learning-enhanced prediction process according to various aspects of this disclosure. Figure 2KIn the prediction phase 267, a recursive forecasting step is employed to generate forecasts for the outcome variable (e.g., sales). This approach can involve using the model’s predictions as inputs for subsequent predictions, enabling the generation of a sequence of forecasts. The outputs are organized into a comprehensive table 269, which can include various relevant variables and their lags. In this example, the table 269 includes date, sales, sales lag 1, sales lag 2, sales lag 3, inventory (Inv), inventory lag 1, gas price (Gas), and gas price lag 1. By incorporating these lagged variables, the adaptive forecasting system 101 can capture temporal dependencies and interactions among different factors, improving the accuracy and robustness of the forecasts. The structured table 269 can provide a clear and detailed view of the predicted outcomes and their influencing factors over the forecast horizon.

[0085] In Figure 2L The production model can be loaded from MLflow. Once the model is loaded, it can generate forecasts using the input data. This process can employ recursive forecasting, where the model’s output at one time step becomes the input for the next time step, creating a sequence of forecasts. Specifically, the model can predict the values for the sales column 271 in the table 269. By filling in the sales column with the forecasted values, the adaptive forecasting system 101 can provide a comprehensive set of predictions that can account for trends in past data and interactions among different variables. The adaptive forecasting system 101 can generate a presentation of the table 269, chart 283, or any other graphical illustration in a user device 104 associated with a user via the visualization module 117. In one example, the user device 104 can include, but is not limited to, any type of mobile terminal, wireless terminal, fixed terminal, or portable terminal. Examples of the user device 104 can include, but are not limited to, a mobile handset, a wireless communication device, a unit, a device, a multimedia computer, a multimedia tablet, an Internet node, a communicator, a desktop computer, a laptop computer, a notebook computer, a netbook computer, a tablet computer, a personal communication system (PCS) device, a personal digital assistant (PDA), an infotainment system, an instrument cluster computer, a television device, or any combination thereof, including accessories and peripherals of these devices or any combination thereof. Moreover, the user device 104 can facilitate various input modes for receiving and generating information, including, but not limited to, touch screen functionality, keyboard and keypad data entry, voice-based input mechanisms, and the like. Any known and future implementations of the user device 104 are also applicable.

[0086] In Figure 2MIn the next stage, once the sales forecast is populated into the table 269, the next stage involves observing the values in the populated sales column, sales lag column, inventory column, inventory lag column, gas price column, and gas price lag column. It is crucial to calculate and populate the lag variables based on the newly forecasted sales values to maintain the integrity of the time series analysis. Specifically, the adaptive forecasting system 101 can calculate sales lag 1, sales lag 2, and sales lag 3, which can represent the sales of the previous one, two, and three time periods, respectively. In addition, the adaptive forecasting system 101 can calculate inventory lag 1 and gas price lag 1, which are the inventory and gas price of the previous time period. These lag values are then populated into the corresponding columns (e.g., 273) of the table 269, ensuring that the dataset remains complete and ready for the subsequent forecast period.

[0087] In the next stage, Figure 2N In the next stage,

[0088] In the next stage, Figure 2O In the next stage,

[0089] In the next stage, Figure 2P In the next stage,

[0090] In the next stage, Figure 2Q and Figure 2RIn one example, once the forecast of the target variable Y is generated for a selected time point, the next step is to populate the lag variables for that forecast. These lag variables (including Sales Lag 1, Sales Lag 2, and Sales Lag 3) are calculated based on the forecast of Y and its historical values. By incorporating these lag variables, the dataset maintains the temporal dependencies necessary for accurate forecasting. Subsequently, these lag variables are used as inputs to the model to generate forecasts for the next time point. Once these forecasts are obtained, they are populated in the remaining cells of the table 269 for Sales, Sales Lag 1, Sales Lag 2, and Sales Lag 3. This is an iterative process that is repeated until the table is filled with forecasts and their lags. The forecast data is written to the data lake in the form of a Spark table. In this example, Figure 2R The table 269 is the final output.

[0091] Figure 2S is a chart illustrating the prediction phase of the machine learning enhanced forecasting process in accordance with various aspects of the present disclosure. The adaptive forecasting system 101 can generate, via the visualization module 117, a comparison chart 283 in the user interface of the user device 104 for evaluating the performance of the prediction model. In one example, the comparison chart 283 can include a line 284 representing the forecasts generated by the prediction model, while line 285 depicts the actual values observed in the real data. Additionally, line 286 represents the forecasts generated by the legacy forecasting system, allowing for a direct comparison between the new forecasting method and the old forecasting method. By comparing these lines, the accuracy and reliability of the forecasting system can be evaluated, with any discrepancies between the forecasted values and the actual values providing valuable feedback for further optimization and improvement of the forecasting process.

[0092] In one example, the comparison chart 283 visualizes not only the forecasted and actual values, but also includes the average percentage difference from the actual values for over 300 products over the past three periods. As the scale of the operation expands and more data (including external factors) is incorporated into the system, improvements quickly accumulate. While the line 284 closely tracks the line 285, indicating that the forecasted values closely align with the actual values, it does not perfectly match, which can be a positive signal as perfect alignment can indicate overfitting. Notably, the line 284 exhibits a closer alignment to the line 285 than the line 286, further highlighting the effectiveness of the new forecasting method in more accurately capturing and predicting trends.

[0093] In one example, the adaptive forecasting system 101 can generate, via the visualization module 117, a presentation of the chart 287 in the user interface of the user device 104 associated with the user (e.g., the user device 104 of the user 102). The chart 287 can include a line 288 representing the forecasted values generated by the prediction model, while line 289 depicts the actual values observed in the real data. Additionally, line 290 represents the forecasts generated by the legacy forecasting system, allowing for a direct comparison between the new forecasting method and the old forecasting method. By comparing these lines, the accuracy and reliability of the forecasting system can be evaluated, with any discrepancies between the forecasted values and the actual values providing valuable feedback for further optimization and improvement of the forecasting process. Figure 2T(As shown). Chart 287 compares the forecast model's estimate 288 with the actual revenue sales volume (RSV) 289. In this example, the y-axis of Chart 287 can represent RSV, ranging from $0 to $14,000,000, and the x-axis of Chart 287 can represent a specified timeline (e.g., from November 2021 to December 2024), providing a clear time-series context for the data points. This visualization allows for detailed analysis of the model's forecast accuracy over time. By juxtaposing the forecast model's estimate 288 with the actual RSV 289, Chart 287 highlights periods of accuracy as well as areas where the model's predictions deviate from the actual values. This comparative analysis helps evaluate the model's performance and identify trends, patterns, and anomalies in sales data over a specified period.

[0094] In one instance, the adaptive prediction system 101 can generate a presentation of chart 290 in the user interface of the user device 104 associated with the user via visualization module 117 (e.g., Figures 2U to 2W (As shown). In this example, the y-axis of Chart 290 can represent actual sales volume in metric tons (MT), quantifying the quantity of product sold from 0 to 2000 metric tons, while the x-axis of Chart 290 can represent a specified timeline tracking sales volume over different time intervals. Chart 290 includes a bar chart 293 indicating actual MT values ​​and compares the forecast 291 of the predictive model with the current situation 292, allowing for an evaluation of the model's performance relative to existing forecasting methods. By juxtaposing the forecast 291 of the predictive model with the current situation 292, Chart 290 can provide insights into how well the model's forecasts match current methods, highlighting areas where it performs better or deviates from current forecasting methods. This comparative analysis can help validate the accuracy and reliability of the model.

[0095] like Figures 2U to 2W As shown, the prediction 291 of the forecasting model touches the bar chart 293, indicating a high degree of agreement between the model's predictions and the actual data points. This demonstrates that the model accurately captures and reflects patterns and trends in the actual data, thus providing reliable predictions. In one instance, the fact that the prediction 291 of the forecasting model touches the bar chart 293 could also indicate a correlation between the model's predictions and the actual values, which can enhance confidence in the model's predictive ability.

[0096] Figure 3 This is a flowchart illustrating the process of adaptive prediction using advanced statistical models and machine learning models according to various aspects of this disclosure. In various embodiments, any of the adaptive prediction system 101 and / or modules 105 to 119 may execute one or more portions of process 300 and use, for example, Figure 5The illustrated chipsets comprising processors and memories are implemented to achieve. Thus, the adaptive forecasting system 101 and / or any of the modules 105-119 can provide the means for implementing various portions of the process 300, as well as the means for implementing embodiments of other processes described herein in conjunction with other components of the system 100. Although the process 300 is illustrated and described as a series of steps, it is contemplated that various embodiments of the process 300 can be performed in any order or combination, and need not include all of the illustrated steps.

[0097] In step 301, the adaptive forecasting system 101 can receive a plurality of data from one or more sources (e.g., the data sources 103). In one instance, the plurality of data can include historical data (e.g., historical sales data), market data, economic indicators, weather data, or any other relevant data.

[0098] In step 303, the adaptive forecasting system 101 can process the plurality of data to select relevant variables. In one instance, the adaptive forecasting system 101 can use imputation techniques to detect and impute missing values in the plurality of data. This can include identifying blanks or anomalies in the dataset where values are missing. Once detected, the adaptive forecasting system 101 can employ various imputation techniques (e.g., mean, median, or mode imputation, forward or backward fill, and K-Nearest Neighbors among advanced algorithms) to fill in these blanks. In one instance, the adaptive forecasting system 101 can perform smoothing and transformation of the relevant variables. Smoothing can include techniques such as moving average or exponential smoothing to reduce noise and volatility in the data, thereby revealing underlying patterns and trends. Transforming the relevant variables can include applying mathematical operations such as scaling, logarithmic transformation, and differencing. These transformations can be crucial for normalizing data, handling outliers, and making time series stationary, which can enhance the ability of the predictive model to efficiently learn from the data. In one instance, the adaptive forecasting system 101 can normalize features in the plurality of data using normalization techniques. Common normalization techniques can include min-max scaling (which can adjust values based on the minimum and maximum values of each feature) and z-score normalization (which can normalize data based on the mean and standard deviation of the data). The adaptive forecasting system 101 can ensure the integrity and consistency of the data by reconciling different formats and structures, in order to create a uniform dataset for subsequent processing and analysis.

[0099] In one example, the relevant variables can include exogenous features, which include economic indicators, weather data, promotional activities, or any time-based data provided by users. In one example, the adaptive forecasting system 101 can utilize feature selection techniques to select relevant variables that have a significant contribution to the performance of the prediction model. The process can include statistical tests, correlation analysis, or machine learning algorithms to evaluate the importance of each variable. By analyzing the relationship between the variables and the target outcome, the adaptive forecasting system 101 can prioritize variables that can provide the strongest predictive power. In one example, the adaptive forecasting system 101 can utilize advanced techniques, such as correlation analysis or principal component analysis (PCA), to evaluate the significance and correlation of each variable. In one example, the adaptive forecasting system 101 can perform simulations to evaluate the significance of one or more variables in a plurality of data. For example, feature importance simulations, including perturbation methods and statistical tests, can be conducted to determine the impact of individual variables on the performance of the model. The adaptive forecasting system 101 can analyze the simulation results to identify variables that have a greater impact on the model predictions.

[0100] In step 305, the adaptive forecasting system 101 can train the prediction model based on the relevant variables and a combination of advanced statistical models and machine learning models. In one example, the advanced statistical models can include ARIMA, SARIMAX, or exponential smoothing. It should be appreciated that the advanced statistical models can include any known or future implementation of time series models. In one example, the machine learning models can include LSTM, Random Forest, deep learning models, Gradient Boosting Machine, Transformer, ExtraTrees, AdaBoost, XGBoost, or LightGBM. It should be appreciated that the machine learning models can include any known or future implementation of time series models.

[0101] In one example, the adaptive forecasting system 101 can identify relevant variables within the plurality of data based on a correlation analysis, which can quantify the strength and direction of the relationship between two continuous variables. In one example, the adaptive forecasting system 101 can identify relevant variables within the plurality of data based on a statistical analysis. In one example, the statistical analysis can include a regression analysis, which can model the relationship between a dependent variable and an independent variable. In one example, the statistical analysis can include a hypothesis test, which can assess the statistical significance of an observed relationship. These relevant variables can have a strong linear relationship with the target variable and can contribute more to explaining the variance of the target variable than randomly generated data, which can lack any inherent relationship with the target variable. The adaptive forecasting system 101 can input the relevant variables into advanced statistical models and machine learning models to analyze the patterns between the relevant variables and the target variable. In one example, the patterns can include temporal patterns, which can examine the behavior of variables over different time periods to identify trends, cycles, or seasonal changes in the target variable. Advanced statistical models can capture temporal series patterns and trends, while machine learning models can learn complex relationships and non-linearities within the data. Integrating statistical models and machine learning models can improve the accuracy and robustness of the forecasts, effectively adapting to various data patterns and external influences.

[0102] In one example, the adaptive forecasting system 101 can use Bayesian optimization to identify the best hyperparameters for the machine learning model. In one example, Bayesian optimization can use a probabilistic model to predict the performance of different hyperparameter configurations and iteratively select the most promising configurations for evaluation. By focusing on the most informative regions of the hyperparameter space, this approach can reduce the number of evaluations required compared to traditional techniques. This process can ensure that the machine learning model is finely tuned to achieve the highest accuracy and performance that adapts to the specific characteristics of the data.

[0103] Additionally or alternatively, the adaptive forecasting system 101 can use a grid search technique to determine the combination of parameters for the advanced statistical model to identify the best configuration that minimizes the prediction error. Grid search can systematically evaluate every possible combination of parameter values, such as the order of autoregression terms (p), the degree of differencing (d), and the order of moving average (q) for an ARIMA model. By training and validating the model on a validation dataset using each parameter set, grid search can identify the combination that can achieve the best performance.

[0104] The adaptive forecasting system 101 can apply cross-validation to evaluate the performance of advanced statistical models (e.g., parametric) and hyperparameter configurations of machine learning models and prevent overfitting. In one example, the models can be trained on some subsets while validated on the remaining subsets, ensuring that each subset is used for validation at least once. This approach can provide a comprehensive evaluation of the model performance under different data partitions, providing insights into their generalization capabilities.

[0105] In one instance, the adaptive forecasting system 101 can utilize parallel processing techniques during the processing of multiple data and training of prediction models. This approach can utilize distributed computing frameworks, for example, distributing data preprocessing tasks (e.g., handling missing values and feature engineering) to multiple processors to reduce the time required for data preparation. Additionally, parallel execution of hyperparameter tuning and model training across various statistical and machine learning models can accelerate the optimization process. The adaptive forecasting system 101 can integrate hyperparameter tuning into machine learning models and parameter optimization into advanced statistical models.

[0106] In step 307, the adaptive forecasting system 101 can evaluate the performance of the trained prediction models based on validation techniques. In one instance, the adaptive forecasting system 101 can rigorously evaluate the accuracy, reliability, and robustness of the prediction models after training. This evaluation can utilize various validation techniques, such as cross-validation or time-series validation, to ensure that the predictive capabilities of the models remain consistent and effective across different datasets or time points. By employing these validation techniques, the system ensures that the trained prediction models meet predefined standards before being deployed for on-demand forecasting.

[0107] In step 309, the adaptive forecasting system 101 can deploy at least one prediction model based on the performance to generate one or more predictions. In one example, the adaptive forecasting system 101 can generate on-demand predictions upon receiving a specific request, allowing users to obtain forecasts based on the latest available data at any given moment. This capability is particularly useful for scenarios requiring immediate insights, such as sudden market changes or emergency inventory management decisions. In another example, the adaptive forecasting system 101 can generate real-time predictions, which can be continuously updated as new data streams in, enabling the system to provide ongoing dynamic forecasts. This real-time capability is crucial for applications requiring continuous monitoring and quick adaptation to changing conditions, such as real-time sales tracking or dynamic supply chain management. In one instance, the adaptive forecasting system 101 can use performance metrics to monitor the performance of at least one deployed model. In one example, the adaptive forecasting system 101 can continuously monitor the performance of deployed prediction models by collecting and analyzing various performance metrics. These performance metrics can include mean absolute error (MAE), root mean squared error (RMSE), mean absolute percentage error (MAPE), or other relevant statistical measures that gauge the accuracy and reliability of model predictions. By tracking these metrics, the adaptive forecasting system 101 can detect deviations or anomalies in model performance, allowing for timely adjustments, retraining, or improvements to the models to adapt to changing patterns and ensure forecasting accuracy.

[0108] In one example, the adaptive forecasting system 101 can train multiple prediction models (e.g., over 300,000 prediction models) within a predetermined time period (e.g., approximately 24 hours) to forecast sales demand for a predetermined number of product-retailer pairs (e.g., 5,000 product-retailer pairs). Leveraging advanced algorithms, the adaptive forecasting system 101 can identify the best model for each individual product-retailer pair by evaluating and comparing the performance of different prediction models. Once the best model is determined, its final predictions are seamlessly integrated into the business systems, ensuring that forecasts are readily available for decision-making processes. This high-throughput and automated approach enables accurate and timely demand forecasting, enhancing inventory management and operational efficiency for retailers.

[0109] One or more implementations disclosed herein include and / or can be implemented using machine learning models. For example, one or more modules of the adaptive forecasting system 101 can be implemented using machine learning models, and / or can be used to train machine learning models. Machine learning models that can be used include, but are not limited to, decision trees, random forests, gradient boosting machines, support vector machines, k-nearest neighbors, neural networks, and / or other machine learning models. Figure 4The training flow diagram 400 in FIG. 4 trains a given machine learning model. The training data 412 can include one or more of stage inputs 414 and known results 418 related to the machine learning model to be trained. The stage inputs 414 can come from any applicable source, including text, visual presentation, data, values, comparisons, stage outputs, for example, from one or more outputs of one or more actions or operations of Figure 3 The known results 418 can be included for machine learning models generated based on supervised or semi-supervised training. Unsupervised machine learning models can not be trained using known results 418. The known results 418 can include known or expected outputs for future inputs that are similar to or belong to the same category as the stage inputs 414 that do not have corresponding known outputs.

[0110] The training data 412 and the training algorithm 420 (e.g., one or more of the modules that implement and / or can be used to train machine learning models) can be provided to a training component 430, which can apply the training data 412 to the training algorithm 420 to generate a machine learning model. According to one implementation, the comparison results 416 comparing previous outputs of the respective machine learning models can be provided to the training component 430 to apply the previous results to retrain the machine learning models. The training component 430 can use the comparison results 416 to update the respective machine learning models. The training algorithm 420 can utilize machine learning networks and / or models, including but not limited to deep learning networks (e.g., deep neural networks (DNNs), convolutional neural networks (CNNs), fully convolutional networks (FCNs), and recurrent neural networks (RNNs)), probabilistic models (e.g., Bayesian networks and graphical models), classifiers (e.g., K-Nearest Neighbors), and / or discriminative models (e.g., decision forests and maximum margin methods), models discussed specifically in the present disclosure, and the like.

[0111] The machine learning models used herein can be trained and / or used by adjusting one or more weights and / or one or more layers of the machine learning model. For example, during training, a given weight can be adjusted (e.g., increased, decreased, removed) based on the training data or input data. Similarly, a layer can be updated, added, or removed based on the training data and / or input data. The final output can be adjusted based on the adjusted weights and / or layers.

[0112] In general, any process or operation discussed in the present disclosure that is understood to be computer-implemented (e.g., the training flow diagram 400) is understood to be implemented using one or more computing devices, such as the computing device 100 of FIG. 1. Figure 3The processes (e.g., the processes illustrated in FIG. 6) can each be performed, for example, by one or more processors of a computer system as described herein. Processes or process steps performed by one or more processors can also be referred to as operations. The one or more processors can be configured to perform such processes by accessing instructions (e.g., software or computer-readable code) that, when executed by the one or more processors, cause the one or more processors to perform the process. The instructions can be stored in a memory of the computer system. The processors can be central processing units (CPUs), graphics processing units (GPUs), or any suitable type of processing units.

[0113] A computer system (e.g., a system or device that implements the processes or operations in the above examples) can include one or more computing devices. The one or more processors of the computer system can be included in a single computing device or distributed among multiple computing devices. The one or more processors of the computer system can be connected to a data storage device. The memory of the computer system can include a respective memory of each of the multiple computing devices.

[0114] Figure 5 An implementation of a computer system that can perform the techniques described herein is shown. The computer system 500 can include a set of instructions that can be executed to cause the computer system 500 to perform any one or more of the methods or computer-based functions disclosed herein. The computer system 500 can operate as a standalone device or can be connected, e.g., using a network, to other computer systems or peripheral devices.

[0115] Unless specifically stated otherwise as apparent from the following discussion, it is appreciated that throughout the description, discussions utilizing terms such as “processing,” “computing,” “calculating,” “determining,” “analyzing” or the like, refer to the action and / or processes of a computer or computing system, or similar electronic computing device, that manipulates and / or transforms data represented as physical quantities (e.g., electronic

[0116] In a similar manner, the term “processor” can refer to any device or portion of a device that manipulates electronic data, e.g., digital data, based on instructions that are stored in memory and that

[0117] In a networked deployment, computer system 500 can operate as a server or as a client user computer in a server-client user network environment, or as a peer-to-peer computer system in a peer-to-peer (or distributed) network environment. Computer system 500 can also be implemented as or incorporated into various devices, such as personal computers (PCs), tablet computers, set-top boxes (STBs), personal digital assistants (PDAs), mobile devices, handheld computers, laptop computers, desktop computers, communication equipment, wireless telephones, landline telephones, control systems, cameras, scanners, fax machines, printers, pagers, personal trusted devices, network equipment, network routers, switches, or bridges, or any other machine capable of executing (sequentially or otherwise) a set of instructions specifying actions to be taken by that machine. In certain implementations, computer system 500 can be implemented using electronic devices that provide voice, video, or data communications. Furthermore, although computer system 500 is shown as a single system, the term "system" should also be considered as including any collection of systems or subsystems that individually or collectively execute one or more sets of instructions to perform one or more computer functions.

[0118] like Figure 5 As shown, computer system 500 may include processor 502, such as a central processing unit (CPU), graphics processing unit (GPU), or both. Processor 502 can be a component in various systems. For example, processor 502 may be part of a standard personal computer or workstation. Processor 502 may be one or more processors, digital signal processors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), servers, networks, digital circuits, analog circuits, combinations thereof, or other devices known now or developed hereafter for analyzing and processing data. Processor 502 may implement software programs, such as manually generated (i.e., programmed) code.

[0119] The computer system 500 can include a memory 504 that can communicate via a bus 508. The memory 504 can be a main memory, a static memory, or a dynamic memory. The memory 504 can include, but is not limited to, computer-readable storage media such as various types of volatile and non-volatile storage media including without limitation random access memory, read-only memory, programmable read-only memory, electrically programmable read-only memory, electrically erasable read-only memory, flash memory, magnetic tape or disk, optical media and the like. In one implementation, the memory 504 includes a cache or random access memory for the processor 502. In alternative implementations, the memory 504 is a separate component in communication with the processor 502, such as a cache memory of a processor, system memory, or other memory. The memory 504 can be an external storage device or database accessible by the processor 502. Examples include a hard disk drive, compact disk ("CD"), digital video disk ("DVD"), memory card, memory stick, floppy disk, universal serial bus ("USB") storage device, or any other device operable to store data. The memory 504 is operable to store instructions executable by the processor 502. The functions, acts or tasks illustrated in the figures or described herein can be performed by instructions stored in the memory 504 and executed by the processor 502. The functions, acts or tasks are independent of the particular type of instruction set, storage media, processor or processing strategy used to execute them. They can be performed by software, in hardware, including

[0120] As shown, the computer system 500 can also include a display 510, such as a liquid crystal display ("LCD"), an organic light emitting diode ("OLED"), a flat panel display, a solid state display, a cathode ray tube ("CRT"), a projector, a printer, or other display device known in the art now or at any time in the future for outputting determined information. The display 510 can serve as an interface for a user to view the functionality of the processor 502, or specifically to interact with software stored in the memory 504 or drive unit 506.

[0121] Additionally or alternatively, the computer system 500 can include an input / output device 512 configured to allow a user to interact with any of the components of the computer system 500. The input / output device 512 can be a digital keyboard, a keyboard, or a cursor control device (such as a mouse or joystick), a touch screen display, a remote control, or any other device operable to interact with the computer system 500.

[0122] The computer system 500 can also or alternatively include a drive unit 506, implemented as a disk drive or optical drive. The drive unit 506 can include a computer-readable medium 522 on which is stored one or more sets of instructions 524, such as software. The instructions 524 can also reside, completely or at least partially, within the main memory 504 and / or within the processing unit 502 during execution thereof by the computer system 500. The main memory 504 and the processing unit 502 also can include, among other things, computer-readable media. The computer-readable medium 522 can also be, or comprise, a tangible medium that stores the set of instructions, such as a set of instructions 524.

[0123] In some systems, the computer-readable medium 522 includes, or is, the set of instructions 524, or receives and executes the set of instructions 524 responsive to a propagated signal, so that a device connected to a network 530 can communicate voice, video, audio, images or any other data over the network 530. Further, the set of instructions 524 can be transmitted or received over the network 530 via the communication port or interface 520 and / or using a bus 508. The communication port or interface 520 can be a part of the processor 502 or can be a separate component. The communication port or interface 520 can be created in software or can be physical connections in hardware. The communication port or interface 520 can be configured to connect with a network 530, external media, the display 510, or any other components in the computer system 500, or combinations thereof. The connection with the network 530 can be a physical connection, such as a wired Ethernet connection or can be established wirelessly as discussed below. Likewise, the additional connections with other components of the computer system 500 can be physical connections or can be established wirelessly. The network 530 can alternatively be directly connected to the bus 508.

[0124] Although the computer-readable medium 522 is shown as a single medium, the term "computer-readable medium" can include a single medium or multiple media, such as a centralized or distributed database, and / or associated caches and servers that store one or more sets of instructions. The term "computer-readable medium" can also include any medium that is capable of storing, encoding or carrying a set of instructions for execution by a processor or that cause a computer system to perform any one or more of the methods or operations disclosed herein. The computer-readable medium 522 can be non-transitory and tangible. The set of instructions 524 can include various commands that instruct the computer system 500 as a processing machine to perform specific operations such as the methods, operations, or processes described herein. The set of instructions 524 can be embodied in various sets of instructions, and various versions of these sets of instructions can be employed as long as they are

[0125] The computer-readable medium 522 can include solid-state memory, such as a memory card or other package that houses one or more non-volatile read-only memories. The computer-readable medium 522 can be random access memory (RAM) or other volatile re-writable memory. Additionally, or alternatively, the computer-readable medium 522 can include a magneto-optical or optical medium, such as a disk or tapes or other storage device used in the encoding, transport, and / or storage of an carrier wave that is a data signal that includes program code and / or data. A digital file attachment to an e-mail or other self-contained information archive or set of archives can be considered a distribution medium that is a tangible storage medium. Accordingly, the disclosure is considered to include any one or more of a computer-readable medium or a distribution medium and other equivalents and successor media, in which data or instructions can be stored.

[0126] In alternative implementations, dedicated hardware implementations, such as application specific integrated circuits, programmable logic arrays and other hardware devices, can be constructed to implement one or more of the methods described herein. Applications that can include the various implementations of apparatuses and systems can broadly include a variety of electronic and computer systems. One or more implementations described herein can implement functions using two or more specific interrelated hardware modules or devices with related control and data signals that can communicate both internally and between external devices on a system. Accordingly, the present systems encompass software, firmware, and hardware implementations.

[0127] The computer system 500 can be connected to a network 530. The network 530 can define one or more networks including wired or wireless networks. The wireless network can be a cellular telephone network, an 802.10, 802.16, 802.20, or WiMAX network. Further, such networks can include public networks such as the Internet, private networks such as an intranet, or a combination of both, and can utilize various current or later-developed network protocols such as, but not limited to, TCP / IP-based protocols. The network 530 can include a wide area network (WAN) such as the Internet, a local area network (LAN), a campus area network, a metropolitan area network, a direct connection such as through a universal serial bus (USB) port, or any other network that can enable data communication. The network 530 can be configured to couple one computing device to another computing device to facilitate the communication of data between these devices. The network 530 can generally enable any of the forms of machine-readable media to be used to transmit information from one device to another. The network 530 can include communication methods by which information can be transmitted between computing devices. The network 530 can be divided into sub-networks. Sub-networks can allow access to all other components connected to it, or sub-networks can limit access between components. The network 530 can be considered a public or private network connection and can include, for example, a virtual private network or encryption or other security mechanisms employed over the public Internet.

[0128] According to various implementations of the present disclosure, the methods described herein can be implemented by software programs that are executable by a computer system. Moreover, in one example non-limiting implementation, implementations can include distributed processing, component / object distributed processing, and parallel processing. Alternatively, virtual computer system processing can be constructed to implement one or more of the methods or functionality as described herein.

[0129] Although this specification describes components and functions implemented in particular implementations, the disclosure is not limited to such components or functions implemented as described herein. For example, one or more of the components described herein could be combined with other components or could be eliminated entirely. Additionally, other components not described herein could be added to the systems and methods described herein. Further, the methods described herein could be implemented by software programs that are executable on programmable computers; this executable program could be stored in any computer readable storage medium, such as storage devices including memories (e.g., RAM, ROM, Flash, etc.), processing devices, and / or computer readable storage media.

[0130] It will be appreciated that, in one embodiment, the steps of the methods discussed are performed by a suitable processor (or processors) of a processing (i.e., computer) system executing instructions (computer readable code) stored in memory. It will also be appreciated that the present disclosure is not limited to any particular implementation or programming technique and that the disclosure can be implemented using any appropriate techniques for implementing the functionality described herein. The present disclosure is not limited to any particular programming language or operating system.

[0131] It should be understood that throughout this specification and the accompanying claims, the words "comprise," "contain," "include," "have," and the like, are taken to specify the presence of stated features, integers, steps, or components but do not preclude the presence or addition of one or more other features, integers, steps, components, or groups thereof. It should also be understood that where the terms "comprise," "comprising," "include," "including," "have," "has," "have," "having," or the like are used in the following description and claims, such terms are intended to be inclusive or open-ended and not exclusive or limiting unless

[0132] Furthermore, although certain embodiments described herein include some features and do not include other features, those of ordinary skill in the art will understand that features from different embodiments can be combined with one another, as the skilled artisan will understand, unless otherwise contraindicated by context or otherwise understood from the specification as a whole. For example, in the claims below, any of the claimed embodiments can be used in any combination.

[0133] Furthermore, some of the embodiments are described herein as a process or a method of implementing a process or method. Therefore, a processor or other means for carrying out such a process or method of implementing the process or method is likewise provided herein. Moreover, any reference to a method, including an object method, is intended to mean a computer- implemented process or method. Furthermore, any reference to an object method is intended to mean a computer- implemented process or method implemented by a processor or other means for carrying out the method.

[0134] In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the application can be practiced without these specific details. In other instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.

[0135] Accordingly, although embodiments of the application have been described herein in relation to particular embodiments thereof, many other changes, modifications and alterations can be made by those skilled in the art upon the reading and understanding of the specification. The present application includes all such modifications and alterations and equivalents thereof. For example, the above described formulas are representative of procedures that can be used. Functions can be added or deleted from the block diagrams and operations can be interchanged between functional blocks. Steps described in the methods described within the scope of the application can be added or deleted.

[0136] The above-disclosed subject matter is to be considered illustrative, and not restrictive, and the appended claims are intended to cover all such modifications, enhancements, and other implementations falling within the true spirit and scope of the disclosure. Thus, to the maximum extent allowed by law, the scope of the present disclosure is to be determined by the broadest permissible interpretation of the following claims and their equivalents, and shall not be restricted or limited to the abovementioned detailed description. While various implementations of the present disclosure have been described above, it should be understood that they have been presented by way of example only, and not limitation. Numerous other implementations can be devised by those skilled in the art, which fall within the scope of the present disclosure. Accordingly, the disclosure is not limited to the above-described implementations, but rather only by the following claims and their equivalents.

Claims

1. A computer-implemented method, comprising: Multiple data sources are received by one or more processors from one or more sources; The plurality of data are processed by the one or more processors to select one or more relevant variables; One or more predictive models are trained by the one or more processors based on the one or more relevant variables and a combination of advanced statistical models and machine learning models; The performance of the trained one or more prediction models is evaluated by the one or more processors based on one or more verification techniques; and The one or more processors deploy at least one prediction model based on the performance to generate one or more predictions.

2. The computer-implemented method according to claim 1, wherein, Processing the multiple data sets to select the one or more relevant variables includes: The one or more processors use one or more interpolation techniques to detect and interpolate missing values ​​in the plurality of data; The one or more processors normalize one or more features of the plurality of data using one or more normalization techniques; and The one or more processors select the one or more relevant variables using one or more feature selection techniques.

3. The computer-implemented method according to claim 2, wherein, Training the one or more prediction models includes: The one or more processors identify the one or more relevant variables within the plurality of data based on one or more of correlation analysis or statistical analysis, wherein the one or more relevant variables have a strong linear relationship with the target variable; and The one or more processors input the one or more relevant variables into the advanced statistical model and machine learning model, wherein the advanced statistical model and machine learning model analyze the pattern between the one or more relevant variables and the target variable.

4. The computer-implemented method according to claim 3 further includes: The one or more processors use Bayesian optimization to identify the optimal hyperparameters of the machine learning model; Cross-validation is applied by the one or more processors to evaluate the performance of one or more hyperparameter configurations and prevent overfitting; and The one or more processors integrate hyperparameter tuning into the machine learning model and parameter optimization into the advanced statistical model.

5. The computer-implemented method according to claim 3, wherein, The advanced statistical models include: Autoregressive Integrated Moving Average (ARIMA), Seasonal Autoregressive Integrated Moving Average with Exogenous Variables (SARIMAX), or Exponential Smoothing.

6. The computer-implemented method according to claim 3, wherein, The machine learning models include: Long Short-Term Memory (LSTM), Random Forest, Deep Learning Model, Gradient Boosting Machine, Transformer, Extreme Random Tree, Adaptive Boosting, XGBoost, or LightGBM.

7. The computer-implemented method according to claim 1, further comprising: The performance of at least one deployed model is monitored by the one or more processors using performance metrics; and The deployed model is retrained by the one or more processors based on updated data to adapt to changing patterns and trends.

8. The computer-implemented method according to claim 1, wherein, The one or more relevant variables include exogenous features, and the exogenous features include economic indicators, weather data, or promotional activities.

9. The computer-implemented method according to claim 1, further comprising: One or more simulations are performed by the one or more processors to evaluate the significance of one or more variables in the plurality of data; and The one or more processors analyze one or more results from the one or more simulations to identify variables that have a significant impact on model predictions.

10. The computer-implemented method according to claim 1, wherein, Parallel processing techniques are used to process the multiple datasets and train the one or more prediction models.

11. A system comprising: One or more processors in a computer system; and At least one non-transitory computer-readable medium storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations including: Receive multiple data from one or more sources; Process the multiple data sets to select one or more relevant variables; One or more predictive models are trained based on the one or more relevant variables and a combination of advanced statistical models and machine learning models. The performance of the trained one or more prediction models is evaluated based on one or more verification techniques; and Based on the performance, deploy at least one prediction model to generate one or more predictions.

12. The system according to claim 11, wherein, Processing the multiple data sets to select the one or more relevant variables includes: One or more interpolation techniques are used to detect and interpolate missing values ​​in the plurality of data; One or more features from the plurality of data are normalized using one or more normalization techniques; and One or more feature selection techniques are used to select the one or more relevant variables.

13. The system according to claim 12, wherein, Training the one or more prediction models includes: Based on one or more correlation analyses or statistical analyses, identify one or more relevant variables within the plurality of data, wherein the one or more relevant variables have a strong linear relationship with the target variable; and The one or more relevant variables are input into the advanced statistical model and the machine learning model, wherein the advanced statistical model and the machine learning model analyze the pattern between the one or more relevant variables and the target variable.

14. The system of claim 13, further comprising: Use Bayesian optimization to identify the optimal hyperparameters of the machine learning model; Cross-validation is applied to evaluate the performance of one or more hyperparameter configurations and to prevent overfitting. and Hyperparameter tuning is integrated into the machine learning model, and parameter optimization is integrated into the advanced statistical model.

15. The system according to claim 13, wherein, The advanced statistical models include: Autoregressive Integrated Moving Average (ARIMA), Seasonal Autoregressive Integrated Moving Average with Exogenous Variables (SARIMAX), or Exponential Smoothing.

16. The system according to claim 13, wherein, The machine learning models include: Long Short-Term Memory (LSTM), Random Forest, Gradient Boosting Machine, Transformer, Extreme Random Tree, Adaptive Boosting, XGBoost, or LightGBM.

17. The system of claim 11, further comprising: Use performance metrics to monitor the performance of at least one deployed model; and The deployed model is retrained based on updated data to adapt to changing patterns and trends.

18. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors of a computing system, cause the one or more processors to perform operations including: Receive multiple data from one or more sources; Process the multiple data sets to select one or more relevant variables; One or more predictive models are trained based on the one or more relevant variables and a combination of advanced statistical models and machine learning models. The performance of the trained one or more prediction models is evaluated based on one or more verification techniques; and Based on the performance, deploy at least one prediction model to generate one or more predictions.

19. The non-transitory computer-readable medium according to claim 18, wherein, Processing the multiple data sets to select the one or more relevant variables includes: One or more interpolation techniques are used to detect and interpolate missing values ​​in the plurality of data; One or more features from the plurality of data are normalized using one or more normalization techniques; and One or more feature selection techniques are used to select the one or more relevant variables.

20. The non-transitory computer-readable medium according to claim 18, wherein, Training the one or more prediction models includes: Based on one or more correlation analyses or statistical analyses, identify one or more relevant variables within the plurality of data, wherein the one or more relevant variables have a strong linear relationship with the target variable; and The one or more relevant variables are input into the advanced statistical model and the machine learning model, wherein the advanced statistical model and the machine learning model analyze the pattern between the one or more relevant variables and the target variable.

Images