Method, program and apparatus for recommending model for demand forecasting

The method addresses the challenge of complex demand forecasting by analyzing data characteristics to recommend suitable models, improving accuracy and reducing training time.

KR1020260162015APending Publication Date: 2026-11-10IMPACTIVE AI INC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
KR1020250058128
Authority / Receiving Office
KR · KR
Patent Type
Applications
Current Assignee / Owner
Filing Date
2025-05-02
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing demand forecasting methods struggle with complex real-world data characteristics, leading to a decline in forecasting performance due to varying dataset sizes, missing values, and temporal patterns, necessitating a systematic approach to recommend suitable models.

Method used

A method that analyzes demand data using multiple feature metrics to generate a recommendation list of models suited to the data characteristics, involving quantitative analysis, model scoring, and neural network classification to optimize model selection.

Benefits of technology

Improves prediction accuracy by recommending optimal models tailored to data characteristics, reducing training time and resource usage while enhancing model selection efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure PAT00002_ABST
    Figure PAT00002_ABST
Patent Text Reader

Abstract

A method, program, and apparatus for demand forecasting using multi-task learning according to one embodiment of the present disclosure are disclosed. The method may include the steps of: acquiring demand data including variables related to temporal order; and analyzing the demand data based on a plurality of feature metrics to generate a recommendation list including at least one model suitable for processing the demand data.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present disclosure relates to a technology for demand forecasting, and specifically to a method and apparatus for recommending a model by considering the characteristics of data for demand forecasting and performing time series forecasting using the recommended model. Background Technology

[0003] Demand forecasting has established itself as an essential skill for operational optimization and strategic decision-making across various industries. Traditionally, demand forecasting has been performed based on statistical techniques or single machine learning models; however, real-world data possesses complex and diverse characteristics, including dataset size, variable structure, the presence of missing values, and temporal patterns. Consequently, single-model approaches frequently experience a sharp decline in forecasting performance when they do not align with the specific characteristics of the data.

[0004] In particular, demand forecasting data often possesses a complex structure involving a mixture of categorical and numerical variables, as well as seasonality, trends, and irregularities based on chronological order. Furthermore, the training efficiency and predictive performance of models can vary significantly depending on data scale and quality (such as missing value ratios or the presence of outliers). Consequently, there is a growing need for technologies that systematically analyze data characteristics and automatically recommend suitable prediction models. Prior art literature

[0006] Korean Patent Publication No. 10-2786927 (March 25, 2025) The problem to be solved

[0007] The present disclosure is conceived in response to the aforementioned background technology and aims to provide a framework that quantitatively analyzes the characteristics of data for demand forecasting and recommends an artificial intelligence model optimized according to each characteristic.

[0008] However, the problems to be solved in this disclosure are not limited to those mentioned above, and other unmentioned problems may be clearly understood based on the description below. means of solving the problem

[0010] A method for recommending a model for demand forecasting, performed by a computing device according to one embodiment of the present disclosure for realizing the aforementioned task, is disclosed. The method may include the step of obtaining demand data including variables related to temporal order; and the step of analyzing the demand data based on a plurality of feature metrics to generate a recommendation list including at least one model suitable for processing the demand data.

[0011] Alternatively, the variables related to the above temporal order may include at least one of date, time, seasonality, time of day, trend, or weather.

[0012] Alternatively, the plurality of characteristic indicators may include a statistical indicator comprising at least one of the sample size, the number of features, the missing value ratio, or the categorical variable ratio; a target analysis indicator comprising at least one of the shape of the data distribution, the skewness of the target variable, and the outlier ratio; and a temporal analysis indicator comprising whether there is a time series pattern.

[0013] Alternatively, the step of generating the recommendation list may include the step of determining the characteristics of the demand data through quantitative analysis of each of the plurality of characteristic indicators.

[0014] Alternatively, the step of generating the recommendation list may further include the step of selecting at least one model that matches each of the characteristics of the determined demand data from among the candidate models pre-set for demand forecasting, and generating independent recommendation lists for each of the plurality of characteristic indicators.

[0015] Alternatively, the step of generating the recommendation list may further include: a step of assigning a score to each of the pre-set candidate models for demand forecasting based on the characteristics of the determined demand data; and a step of integrating the assigned scores to determine the top N (N is a natural number) of the candidate models as recommended models.

[0016] Alternatively, the step of generating the recommendation list may further include: a step of converting the characteristics of the determined demand data into embedding vectors; and a step of inputting the converted embedding vectors into a pre-trained neural network model to classify M (where M is a natural number) models among the candidate models as recommendation models.

[0017] Alternatively, the method may further include the step of inputting the demand data into at least one recommended model to generate time series data for demand forecasting.

[0018] A computer program stored on a computer-readable storage medium is disclosed in accordance with one embodiment of the present disclosure for realizing the aforementioned objectives. When the computer program is executed on one or more processors, it performs operations for recommending a model for demand forecasting. At this time, the operations may include: an operation of acquiring demand data including variables related to temporal order; and an operation of analyzing the demand data based on a plurality of characteristic indicators to generate a recommendation list including at least one model suitable for processing the demand data.

[0019] According to one embodiment of the present disclosure for realizing the aforementioned objectives, a computing device for recommending a model for demand forecasting is disclosed. The device may include a processor comprising at least one core; a memory comprising program codes executable on the processor; and a network unit for acquiring demand data comprising variables related to temporal order. In this case, the processor may analyze the demand data based on a plurality of characteristic indicators and generate a recommendation list comprising at least one model suitable for processing the demand data. Effects of the invention

[0021] The present disclosure can improve prediction accuracy compared to random or empirical model selection by recommending an optimal demand forecasting model suited to data characteristics. Furthermore, the present disclosure provides an environment in which users can quickly select an appropriate model without the need for unnecessary overuse of model search or hyperparameter tuning, thereby reducing training time for model construction and optimizing computing resources. Brief explanation of the drawing

[0023] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure. FIGS. 2 to 4 are conceptual diagrams illustrating a list generation process for model recommendation according to one embodiment of the present disclosure. FIG. 5 is a flowchart illustrating a model recommendation method for demand forecasting according to one embodiment of the present disclosure. Specific details for implementing the invention

[0024] Embodiments of the present disclosure are described below with reference to the attached drawings so that those skilled in the art (hereinafter, those skilled in the art) can easily implement them. The embodiments presented in the present disclosure are provided to enable those skilled in the art to use or implement the contents of the present disclosure. Accordingly, various modifications to the embodiments of the present disclosure will be apparent to those skilled in the art. That is, the present disclosure may be embodied in various different forms and is not limited to the embodiments below.

[0025] Throughout the specification of the present disclosure, identical or similar reference numerals refer to identical or similar components. Additionally, to clearly explain the present disclosure, reference numerals in the drawings that are unrelated to the description of the present disclosure may be omitted.

[0026] The term “or” as used in this disclosure is intended to mean an implicit “or” rather than an exclusive “or.” That is, unless otherwise specified in this disclosure or its meaning is not clear from the context, “X uses A or B” should be understood to mean one of the natural implicit substitutions. For example, unless otherwise specified in this disclosure or its meaning is not clear from the context, “X uses A or B” may be interpreted as X using A, X using B, or X using both A and B.

[0027] The term “and / or” as used in this disclosure should be understood to refer to and include all possible combinations of one or more of the enumerated related concepts.

[0028] The terms “comprising” and / or “comprising” as used in this disclosure should be understood to mean the presence of certain features and / or components. However, the terms “comprising” and / or “comprising” should be understood not to exclude the presence or addition of one or more other features, other components and / or combinations thereof.

[0029] Where not otherwise specified in the present disclosure or where it is not clear from the context that the singular form indicates, the singular should generally be interpreted as including “one or more.”

[0030] The term “the N (N is a natural number)” used in this disclosure may be understood as an expression used to distinguish the components of this disclosure from one another according to certain criteria, such as functional perspectives, structural perspectives, or convenience of explanation. For example, components performing different functional roles in this disclosure may be distinguished as a first component or a second component. However, components that are substantially identical within the technical scope of this disclosure but must be distinguished for the convenience of explanation may also be distinguished as a first component or a second component.

[0031] The term “connection” as used in the present disclosure should be interpreted to include not only cases where the components are “directly connected,” but also cases where other components are “present” in between, and cases where they are “electrically connected” with other components in between.

[0032] The term “acquisition” as used in this disclosure may be understood to refer not only to receiving data through a wireless communication network with an external device or system, but also to generating or receiving data in an on-device form.

[0033] Meanwhile, the terms "module" or "unit" used in this disclosure may be understood as referring to an independent functional unit that processes computing resources, such as a computer-related entity, firmware, software or a part thereof, hardware or a part thereof, or a combination of software and hardware. In this case, "module" or "unit" may be a unit composed of a single element, or a unit expressed as a combination or set of multiple elements. For example, in a narrow sense, "module" or "unit" may refer to a hardware element of a computing device or a set thereof, an application program that performs a specific function of software, a procedure implemented through software execution, or a set of instructions for program execution. Furthermore, in a broad sense, "module" or "unit" may refer to the computing device itself that constitutes the system, or an application executed on the computing device. However, since the above-described concept is merely an example, the concept of "module" or "part" may be defined in various ways within the scope understandable to those skilled in the art based on the contents of this disclosure.

[0034] As used in this disclosure, the term “model” may be understood as a system implemented using mathematical concepts and language to solve a specific problem, a set of software units to solve a specific problem, or an abstract model regarding a process to solve a specific problem. For example, a machine learning “model” may refer to an overall system that possesses problem-solving capabilities through learning based on data.

[0035] The explanation of the foregoing terms is intended to aid in understanding the present disclosure. Accordingly, it should be noted that unless a foregoing term is explicitly stated as a matter limiting the content of the present disclosure, it is not to be used in the sense of limiting the technical concept of the content of the present disclosure.

[0036] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure.

[0037] A computing device (100) according to one embodiment of the present disclosure may be a hardware device or a part of a hardware device that performs comprehensive processing and computation of data, or it may be a software-based computing environment connected to a communication network. For example, the computing device (100) may be a server that performs intensive data processing functions and shares resources through communication with clients. Additionally, the computing device (100) may be a cloud system that enables multiple servers and clients to process data comprehensively. Since the above description is merely one example regarding the type of computing device (100), the type of computing device (100) may be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0038] Referring to FIG. 1, a computing device (100) according to one embodiment of the present disclosure may include a processor (110), a memory (120), and a network unit (130). However, since FIG. 1 is merely an example, the computing device (100) may include other configurations for implementing a computing environment. Additionally, only some of the disclosed configurations may be included in the computing device (100).

[0039] A processor (110) according to one embodiment of the present disclosure may be understood as a constituent unit comprising hardware and / or software for performing computing operations. For example, the processor (110) may process instructions generated as a result of user interaction through a user interface. Additionally, the processor (110) may read a computer program and perform data processing for machine learning. The processor (110) may process computational processes such as processing input data for machine learning, feature extraction for machine learning, and error calculation based on backpropagation. A processor (110) for performing such data processing and operations may include a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), a tensor processing unit (TPU), an application specific integrated circuit (ASIC), or a field programmable gate array (FPGA). Since the above-described type of processor (110) is merely an example, the type of processor (110) can be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0040] The processor (110) can analyze demand data including variables related to temporal order based on multiple feature metrics. The processor (110) can derive quantitative result values ​​by analyzing the demand data for each metric. The processor (110) can determine the characteristics of the demand data for each metric by analyzing the quantitative result values ​​derived for each metric according to the comparison logic provided for each metric. For example, the processor (110) can derive a numerical value representing the size of the demand data by analyzing the demand data based on a metric called data size. Then, the processor (110) can determine the size of the demand data as small, medium, or large by comparing the numerical value representing the size of the demand data with threshold values ​​for classifying the data size as small, medium, or large. The feature metrics can be composed of various types, including data size, missing value ratio, and distribution shape. Since independent threshold values ​​and comparison criteria are set for each of these feature metrics, the processor (110) can determine the characteristics of the demand data for each metric by analyzing the metrics individually.

[0041] The processor (110) can determine at least one model suitable for analyzing demand data based on the analysis results performed based on multiple characteristic indicators. Once the characteristics of the demand data are determined through the preceding analysis, the processor (110) can select at least one model that matches the characteristics of the demand data from among the pre-set candidate models for demand forecasting. At this time, the selection process can be performed by searching a database of pre-matched candidate models according to the degree of suitability for each characteristic indicator. The selection process can also be performed by calculating a score indicating how suitable the pre-set candidate models are based on the characteristics of the demand data. Meanwhile, the selection process can be performed based on deep learning-based computation as well as rule-based computation. The processor (110) can predict an optimal demand forecasting model that matches the characteristics of the demand data by inputting the characteristics of the demand data into a neural network model based on meta-learning. At this time, the optimal demand forecasting model may be at least one of the pre-set candidate models.

[0042] When at least one model suitable for analyzing demand data is determined, the processor (110) can generate a recommendation list containing at least one model. The processor (110) can determine a model to be used for demand forecasting from the recommendation list based on a user command. Then, the processor (110) can perform training on the model to be used for demand forecasting. For example, if linear regression is included in the recommendation list because the distribution of the demand data is linear, the processor (110) can perform training on linear regression based on a user command to decide to use linear regression. Specifically, the processor (110) can generate time series data for forecasting future demand by inputting training data containing demand data for which characteristic analysis has been completed into a linear regression model. The processor (110) can compare the time series data generated by the linear regression model with labels containing actual values ​​of intervals defined as future points in time using a loss function. At this time, the loss function may include at least one of means squared error or cross-entropy. The processor (110) can update the parameters constituting the neural network of the linear regression model through backpropagation using the loss value generated through this comparison. This computational process can be performed repeatedly over several epochs.

[0043] The processor (110) can generate time series data for predicting future demand by analyzing demand data based on a model that has been trained as described above. Depending on the demand data of the application field on which training was performed, the demand forecasting of the present disclosure can be flexibly applied to various industries and demand forecasting scenarios such as [Table 1] below. For example, the processor (110) can recommend at least one model by analyzing the characteristics of product shipment data that include variables related to temporal order. Then, the processor (110) can determine the model selected by user command based on the recommendation as the model for predicting product shipment volume and perform training. The processor (110) can generate time series data for predicting the future shipment volume of a specific product by inputting past shipment data of a specific product into the trained model. At this time, the time series data may be categorical and numerical data that can verify changes in shipment volume from the past to a specific point in the future of a specific product.

[0044] Prediction problem applied area Product shipment volume forecast - Production planning and logistics management in manufacturing - Sales trend analysis and seasonal shipment optimization - Coordination of logistics flow within the supply chain Inventory forecast - Prevention of excess and shortages - Setting and managing safety stock levels - Optimization of warehouse and logistics space Energy Demand Forecast - Forecasting electricity, gas, and oil consumption - Forecasting fluctuations in renewable energy (solar, wind, etc.) production - Optimizing energy supply and distribution Raw material requirements forecast - Establishing raw material purchasing plans for the manufacturing industry - Optimizing production processes and reducing costs - Establishing raw material securing strategies for supply chain stability Raw material price forecast - Managing raw material market price volatility - Optimizing long-term contract and purchase timing - Forecasting global resource extraction and trade costs

[0045] Meanwhile, since [Table 1] is merely an example, the demand forecasting problem of the present disclosure can be expanded in various ways based on the content disclosed in [Table 1]. A memory (120) according to one embodiment of the present disclosure can be understood as a configuration unit including hardware and / or software for storing and managing data processed by a computing device (100). That is, the memory (120) can store data of any form generated or determined by the processor (110) and data of any form received by the network unit (130). For example, the memory (120) may include at least one type of storage medium among flash memory type, hard disk type, multimedia card micro type, card type memory, RAM (random access memory), SRAM (static random access memory), ROM (read-only memory), EEPROM (electrically erasable programmable read-only memory), PROM (programmable read-only memory), magnetic memory, magnetic disk, or optical disk. Additionally, the memory (120) may include a database system that controls and manages data in a predetermined system. Since the above-described types of memory (120) are merely examples, the types of memory (120) may be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0046] Memory (120) can structure and organize data, combinations of data, and program code executable by the processor (110) that are necessary for the processor (110) to perform calculations. For example, memory (120) can store program code that causes the processor (110) to process data, program code that causes the processor (110) to create a machine learning model and perform demand forecasting calculations using the created machine learning model, and various data generated as the program code is executed.

[0047] A network unit (130) according to one embodiment of the present disclosure may be understood as a configuration unit that transmits and receives data through any known form of wired or wireless communication system. For example, the network unit (130) may perform data transmission and reception using wired or wireless communication systems such as a local area network (LAN), wideband code division multiple access (WCDMA), long term evolution (LTE), wireless broadband internet (WiBro), 5th generation mobile communication (5G), ultra-wide-band wireless communication, ZigBee, radio frequency (RF) communication, wireless LAN, wireless fidelity (Wi-Fi), near field communication (NFC), or Bluetooth. Since the communication systems described above are merely examples, wired or wireless communication systems for data transmission and reception of the network unit (130) may be applied in various ways other than those described above.

[0048] The network unit (130) can receive data necessary for the processor (110) to perform calculations through wired or wireless communication with any system or any client. Additionally, the network unit (130) can transmit data generated through the processor (110)'s calculations through wired or wireless communication with any system or any client. The network unit (130) can transmit and receive data generated through interaction with an entity through communication with any entity. For example, the network unit (130) can obtain demand data including variables indicating temporal order through communication with a database. Furthermore, the network unit (130) can transmit data processed by the processor (110) to a database or a client for verifying demand forecast results.

[0049] FIGS. 2 to 4 are conceptual diagrams illustrating a list generation process for model recommendation according to an embodiment of the present disclosure. FIGS. 2 and 3 illustrate a list generation process performed through rule-based operations of a computing device (100) according to an embodiment of the present disclosure, and FIG. 4 illustrates a list generation process performed through deep learning-based operations of a computing device (100) according to an embodiment of the present disclosure.

[0050] Referring to FIGS. 2 to 4, the process (S110) of performing characteristic indicator analysis based on demand data (11) during the process of generating a list for model recommendation can be performed commonly in both rule-based operations and deep learning-based operations.

[0051] Variables related to the temporal order included in the demand data (11) may include at least one of date, time, seasonality, time of day, trend, or weather. Date or time may include year, month, day, day of the week, quarter, whether it is a public holiday, whether it is a weekend, etc. Seasonality may include seasons such as spring, summer, autumn, and winter, as well as seasons when specific events such as Black Friday occur. Time of day may include hour, minute, AM, PM, night, etc. Trend may indicate a pattern of rising or falling over time. Weather may include temperature, precipitation, humidity, wind speed, etc. The demand data (11) may be configured in at least one form of categorical or numerical, including these variables related to the temporal order.

[0052] A computing device (100) can determine the characteristics of demand data (11) through quantitative analysis of each of a plurality of characteristic indicators (S110). The plurality of characteristic indicators may include a statistical indicator (13) including at least one of the number of samples, the number of features, the ratio of missing values, or the ratio of categorical variables; a target analysis indicator (15) including at least one of the shape of the data distribution, the skewness of the target variable, and the ratio of outliers; and a temporal analysis indicator (17) including whether there is a time series pattern. The computing device (100) can determine the characteristics of demand data (11) for each indicator by individually analyzing each of the statistical indicator (13), the target analysis indicator (15), and the temporal analysis indicator (17).

[0053] For example, the sample size is a characteristic regarding the size of the data and can be calculated as the number of rows of the data. The feature count is a characteristic regarding the dimension of the data and can be calculated as the number of columns of the data. The missing value ratio is the ratio of missing values ​​included in the total data and can be calculated as the average value of the entire data. The ratio of categorical variables can be determined based on whether the data type is categorized or whether the number of unique values ​​is small. The shape of the data distribution is a characteristic regarding whether the features of the data have a linear or non-linear structure and can be determined by calculating the Pearson correlation coefficient for numerical features. The skewness of the target variable is a characteristic indicating the bias of the target variable and can be calculated based on the mean and standard deviation of the target variable. The presence of a time series pattern is an analysis of time dependency and can be calculated based on the autocorrelation function. The computing device (100) can determine the characteristics of the demand data (11) by analyzing the demand data by indicator through an indicator-based calculation method such as the example described above. The criteria for determining the characteristics of the demand data (11) by indicator can be summarized as shown in [Table 2] below.

[0054] Characteristic indicators Judgment criteria Number of samples Small scale (sample size < 10k) Medium scale (10k ≤ sample size ≤ 1M) Large scale (sample size > 1M) Number of features Low dimensionality (number of features < 30) High dimension (number of features ≥ 30) Missing Value Ratio Many missing values ​​(> 10%) Few missing values ​​(≤ 10%) Proportion of categorical variables Many categorical variables (> 40%) Few categorical variables (≤ 40%) Shape of data distribution linear relationship non-linear relationship Skewness of the target variable Severe bias (skewness > 1) normal distribution Outlier ratio Many outliers (> 5%) Whether there is a temporal pattern Time dependency exists Non-existence of time dependency

[0055] Referring to [Table 2], if the number of samples of demand data (11) exceeds 1M, the computing device (100) can determine the characteristics regarding the size of the demand data (11) on a large scale. In this way, the computing device (100) can quantitatively analyze the demand data (11) for each indicator and determine the characteristics for each indicator by comparing them with the judgment criteria. Meanwhile, the specific numerical values ​​disclosed in [Table 2] regarding the judgment criteria are merely examples, so the present disclosure is not limited thereto. Referring to FIG. 2, the computing device (100) can select at least one model that matches each of the characteristics of the demand data (11) determined through S110 from among the pre-set candidate models for demand forecasting (S120). Then, the computing device (100) can generate independent recommendation lists (19) for each of the multiple characteristic indicators. At this time, the pre-set candidate models may be models whose utility in the field of demand forecasting has been pre-verified. The computing device (100) can store information regarding recommended models by matching a candidate model suitable for processing to each characteristic indicator based on the individual characteristics of the candidate models. For example, pre-validated models that can be set as candidate models are as shown in [Table 3] below. And, information regarding recommended models matched to each characteristic indicator is as shown in [Table 4] below.

[0056] Candidate Model characteristic Applicability Linear Regression Learns the linear relationship between variables. Simple yet highly interpretable, making it useful for small datasets. When simple pattern learning and interpretation are important. Random Forest It learns non-linear relationships in data and can evaluate variable importance. Advantageous for processing complex patterns. A problem dealing with non-linear relationships and variable importance. XGBoost Provides high predictive performance through boosting techniques. Includes regularization to prevent overfitting. Problems requiring small datasets and high predictive performance. LightGBM (LGBM) Similar to XGBoost but faster and more memory-efficient. Strong in processing large datasets. Problems requiring large-scale dataset processing and rapid learning. CatBoost Optimized for handling categorical variables. Allows the use of categorical data without separate preprocessing. Optimized for problems involving categorical data. Feed Forward Neural Network Learning complex data patterns through multiple layers. Can be utilized by embedding continuous and categorical data. Learning complex data patterns and neural network-based prediction. Deep Tabular Networks A neural network model designed to learn from tabular data. It combines the advantages of Transformers, MLPs, and tree-based learning techniques. Useful when handling various data structures and problem types. TabNet Transformer-based design that enables interpretable decision-making processes. Delivers high performance on tabular data. Problems requiring interpretability and high predictive performance. SAINT (Self-Attention and Intersample Transformers) A transformer-based model for processing tabular data, supporting learning of interactions between samples. Learning interactions between tabular data and utilizing advanced transformer structures. FT-Transformer Transformer-based architecture for simultaneously processing categorical and numerical variables. Individually embeds each feature using a Feature Tokenizer and effectively learns interactions between features using Self-Attention. Problems requiring high-performance prediction on large datasets involving complex feature interactions and a mix of categorical and numerical data.

[0057] Characteristic indicators Judgment criteria Recommended Model Number of samples Small scale (sample size < 10k) Linear Regression, Random Forest, CatBoost Medium scale (10k ≤ sample size ≤ 1M) XGBoost, CatBoost, LightGBM, TabNet, SAINT, FT-Transformer Large scale (sample size > 1M) LightGBM, TabNet, SAINT, FT-Transformer Number of features Low dimensionality (number of features < 30) Linear Regression, Random Forest, CatBoost High dimension (number of features ≥ 30) XGBoost, LightGBM, TabNet, SAINT, FT-Transformer Missing Value Ratio Many missing values ​​(> 10%) CatBoost, Random Forest, LightGBM Few missing values ​​(≤ 10%) All candidate models are possible (especially LightGBM, XGBoost, TabNet, FT-Transformer) Proportion of categorical variables Many categorical variables (> 40%) CatBoost, LightGBM, SAINT, FT-Transformer Few categorical variables (≤ 40%) XGBoost, TabNet Shape of data distribution linear relationship TabNet, SAINT, LightGBM, FT-Transformer non-linear relationship All candidate models are possible Skewness of the target variable Severe bias (skewness > 1) TabNet, SAINT, FT-Transformer normal distribution All candidate models are possible Outlier ratio Many outliers (> 5%) Random Forest, CatBoost, TabNet Whether there is a temporal pattern Time dependency exists TabNet, SAINT, FT-Transformer Non-existence of time dependency All candidate models are possible

[0058] The computing device (100) can manage candidate models and characteristic indicators by matching them as shown in [Table 4]. If the characteristics of the demand data (11) related to the number of samples are determined on a large scale, the computing device (100) can select at least one of LightGBM, TabNet, SAINT, and FT-Transformer as a recommended model related to the number of samples according to [Table 4]. Additionally, if the characteristics of the demand data (11) related to the number of features are determined on a high scale, the computing device (100) can select at least one of XGBoost, LightGBM, TabNet, SAINT, and FT-Transformer as a recommended model related to the number of features according to [Table 4]. In this way, the computing device (100) can select a recommended model for each of the characteristics of the demand data (11). Furthermore, the computing device (100) can generate a list of recommended models independent of each indicator. When a list is organized independently by indicator in this way, it can help the user easily identify which model is suitable according to the characteristics of the demand data and make appropriate decisions. Referring to FIG. 3, the computing device (100) can assign a score to each of the pre-set candidate models for demand forecasting based on the characteristics of the demand data (11) determined through S110 (S210). The computing device (100) can calculate a score indicating how well each candidate model fits the characteristics of the demand data (11). For example, if the characteristics related to the sample size among the characteristics of the demand data (11) are determined to be small, the computing device (100) can assign 3 points to the linear regression model, 2 points to the random forest, 1 point to XGBoost, 3 points to CatBoost, 1 point to LightGBM, and 0 points to the remaining models. Such score assignment is determined based on the individual characteristics of the models.That is, linear regression models operate well regardless of the size of the data, and models such as XGBoost and LightGBM may not be suitable for processing small-scale data because they can cause problems such as overfitting in small-scale data. The computing device (100) can evaluate candidate models by assigning a score to each candidate model according to the characteristics of the demand data (11) based on these characteristics of the candidate models.

[0059] And, the computing device (100) can select the top N models (N is a natural number) among the candidate models as recommended models by integrating the scores assigned to each candidate model according to the characteristics of the demand data (11) (S220). For example, if the characteristics of the demand data are determined to be small, have many missing values, have a non-linear form, have many categorical variables, have severe bias, and have time dependency, the computing device (100) can evaluate the ranking by summing the scores for each of the six characteristics for each candidate model such as linear regression, random forest, XGBoost, etc. If the sum score of linear regression is 4 points, the sum score of random forest is 12 points, and the sum score of XGBoost is 9 points, the ranking of the three models can be evaluated as random forest, XGBoost, and linear regression. The computing device (100) can select the top N models from the overall ranking as recommended models according to the number of selections N set according to user commands. This method can determine the overall characteristics of the demand data (11) comprehensively, so that accurate results can be obtained with minimal user intervention.

[0060] Referring to FIG. 4, the computing device (100) can convert the characteristics of the demand data (11) determined through S110 into an embedding vector (30) (S310). The computing device (100) can individually convert the characteristics of the demand data (11) into an embedding vector that can be analyzed by a neural network, and can combine the converted embedding vectors to create a single embedding vector (30).

[0061] The computing device (100) inputs the embedding vector (30) transformed via S310 into a pre-trained neural network model (40) to classify M (M is a natural number) of candidate models as recommended models (S320). Then, the computing device (100) can synthesize the classification results to generate a recommendation list (50). The neural network model (40) is a multi-classification model that receives the embedding vector (30) as input and classifies recommended model classes, and may be any one of models with a low risk of initial overfitting, such as Random Forest or Gradient Boosting. The neural network model (40) can generate recommended model classes by understanding the relationship between the characteristics of the demand data (11) and the candidate models based on the embedding vector (30). During the learning process, the recommended model class, which is the output of the neural network model (40), can be compared with a label. In addition, the loss value calculated through a loss function such as cross-entropy during the comparison process can be used to update the parameters of the neural network model (40). This method can produce results with sufficient accuracy in an environment rich in demand data (11), can flexibly process various types of demand data (11), and can reliably automate all computation processes.

[0062] Meanwhile, a computing device (100) according to one embodiment of the present disclosure may use a combination of the three list generation methods described in FIGS. 2 to 4. For example, the computing device (100) may generate a first recommendation list including a recommendation model for each characteristic of the demand data (11) and a recommendation model that considers all characteristics by using both the rule-based methods of FIGS. 2 and FIG. 3. Then, the computing device (100) may generate a second recommendation list including a comprehensive judgment result regarding the characteristics of the demand data (11) by using the deep learning-based method of FIG. 4 based on user feedback regarding the first recommendation list. Since such a hybrid method can mutually compensate for the weaknesses of the rule-based method and the deep learning-based method, the accuracy and reliability of the recommendation results can be improved.

[0063] FIG. 5 is a flowchart illustrating a model recommendation method for demand forecasting according to one embodiment of the present disclosure.

[0064] Referring to FIG. 5, a computing device (100) according to one embodiment of the present disclosure can acquire demand data including variables related to temporal order (S410). For example, the computing device (100) can collect demand data including variables related to temporal order through communication with a database for each demand forecasting field. The computing device (100) can construct a data set for each demand forecasting field by sorting the demand data according to temporal order. At this time, the demand data included in the data set for each demand forecasting field can be subdivided and distinguished according to variables related to temporal order.

[0065] The computing device (100) can analyze demand data based on a plurality of characteristic indicators and generate a recommendation list including at least one model suitable for processing demand data (S420). The specific computation process for generating the recommendation list will be replaced by the description of FIGS. 2 to 4.

[0066] The computing device (100) can generate time series data for demand forecasting by inputting demand data into at least one model recommended via S420. Once the final model selection is completed via the recommendation via S420 and user command, the computing device (100) can perform training on the selected model. Then, the computing device (100) can generate time series data for demand forecasting based on newly input demand data using the trained model.

[0067] The various embodiments of the present disclosure described above may be combined with additional embodiments and modified to the extent understandable to those skilled in the art in light of the detailed description above. The embodiments of the present disclosure are illustrative in all respects and should be understood as not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form. Accordingly, all modifications or variations derived from the meaning, scope, and equivalents of the claims of the present disclosure should be interpreted as being included within the scope of the present disclosure.

Claims

Claim 1 A method for recommending a model for demand forecasting, performed by a computing device including at least one processor, comprising: a step of acquiring demand data including variables related to temporal order; and a step of analyzing the demand data based on a plurality of feature metrics to generate a recommendation list including at least one model suitable for processing the demand data. Claim 2 A method according to claim 1, wherein the variable related to the temporal order comprises at least one of date, time, seasonality, time of day, trend, or weather. Claim 3 A method according to claim 1, wherein the plurality of characteristic indicators include: a statistical indicator comprising at least one of the number of samples, the number of features, the ratio of missing values, or the ratio of categorical variables; a target analysis indicator comprising at least one of the shape of the data distribution, the skewness of the target variable, and the ratio of outliers; and a temporal analysis indicator comprising whether there is a time series pattern. Claim 4 A method according to claim 1, wherein the step of generating the recommendation list comprises the step of determining the characteristics of the demand data through quantitative analysis of each of the plurality of characteristic indicators. Claim 5 In claim 4, the step of generating the recommendation list further comprises the step of selecting at least one model that matches each of the characteristics of the determined demand data from among the candidate models pre-set for demand forecasting, and generating an independent recommendation list for each of the plurality of characteristic indicators. Claim 6 In claim 4, the step of generating the recommendation list further comprises: a step of assigning a score to each of the pre-set candidate models for demand forecasting based on the characteristics of the determined demand data; and a step of integrating the assigned scores to determine the top N (N is a natural number) of the candidate models as recommended models. Claim 7 In claim 4, the step of generating the recommendation list further comprises: a step of converting the characteristics of the determined demand data into embedding vectors; and a step of inputting the converted embedding vectors into a pre-trained neural network model to classify M (M is a natural number) models among the candidate models as recommendation models. Claim 8 A method according to claim 1, further comprising the step of inputting the demand data into at least one recommended model to generate time series data for demand forecasting. Claim 9 A computer program stored on a computer-readable storage medium, wherein, when the computer program is executed on one or more processors, the computer program performs operations for recommending a model for demand forecasting, the operations comprising: an operation of acquiring demand data including variables related to temporal order; and an operation of analyzing the demand data based on a plurality of feature metrics to generate a recommendation list including at least one model suitable for processing the demand data. Claim 10 A computing device for recommending a model for demand forecasting, comprising: a processor including at least one core; a memory including program codes executable on the processor; and a network unit for acquiring demand data including variables related to temporal order; wherein the processor analyzes the demand data based on a plurality of feature metrics and generates a recommendation list including at least one model suitable for processing the demand data.