Dynamic performance assessment adjustment system and method based on machine learning, and storage medium

Abstract

The invention provides a dynamic performance assessment adjustment system and method based on machine learning, and a storage medium. The system comprises a data acquisition and processing module, a feature extraction and model construction module, a big data analysis and real-time adjustment module, and a system monitoring and maintenance module. The dynamic performance appraisal adjustment method comprises the following steps: collecting and primarily processing original data; extracting features from the preliminarily processed data through a machine learning algorithm and constructing a performance assessment prediction model; performing deep analysis on the preliminarily processed data by using a prediction model, and dynamically adjusting a performance assessment standard; further dynamically optimizing and updating a prediction model according to an analysis result to adapt to environment change and new data input; the big data analysis and real-time adjustment module carries out automatic decision making to further improve the response speed and efficiency of the system; and the system monitoring and maintenance module tracks and evaluates various performance indexes of the system in real time. According to the invention, the scientificity, the accuracy and the efficiency of enterprise performance management can be obviously improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning, and in particular relates to a dynamic performance appraisal adjustment system, method and storage medium based on machine learning. Background Art

[0002] Existing performance appraisal systems are usually based on static indicators and manual evaluation, using a standardized scoring system to regularly evaluate employee performance. These evaluations often rely on preset goals, historical data, and the subjective judgment of supervisors. Common evaluation methods include:

[0003] 1. Quantitative assessment: scoring based on indicators such as sales volume and number of completed tasks.

[0004] 2. Qualitative assessment: Evaluate employee performance through supervisor evaluation, colleague feedback, etc.

[0005] 3. Annual evaluation: A comprehensive performance evaluation is generally conducted once a year, and the evaluation results are often used for salary adjustments and promotion decisions.

[0006] Although the existing appraisal methods can evaluate employee performance to a certain extent, they still have the following major flaws:

[0007] 1. Lack of real-time performance: The existing assessment system is often based on periodic evaluation, which cannot reflect employees' real-time performance and market dynamics in a timely manner, resulting in delayed performance feedback.

[0008] 2. Poor adaptability: In a rapidly changing market environment, traditional assessment standards are often unable to be flexibly adjusted, resulting in assessment results that cannot truly reflect the contributions of employees.

[0009] 3. Insufficient data utilization: Existing systems often fail to fully tap and utilize the large amount of available data, and the decision-making process relies on subjective judgment, resulting in a lack of scientificity and rationality in performance management.

[0010] 4. Inconsistent goals: Due to the lack of unified assessment standards between departments and projects, there may be conflicts between different goals, affecting the overall performance management effect.

[0011] 5. Inefficiency: The manual evaluation process is cumbersome and time-consuming, prone to errors, and affects the fairness and objectivity of the assessment.

[0012] In view of the above-mentioned defects, the present invention designs a dynamic performance appraisal adjustment system, method and storage medium based on machine learning. Summary of the invention

[0013] In view of the shortcomings in the prior art, the present invention provides a dynamic performance appraisal adjustment system, method and storage medium based on machine learning. Through real-time data analysis and intelligent algorithms, real-time performance monitoring and feedback are achieved (building a system that can monitor employee performance in real time and provide timely feedback, so that the appraisal results are more timely), dynamic adjustment of appraisal standards (realizing automatic adjustment of appraisal standards to ensure that they can adapt to market changes and project requirements, and improving the flexibility of performance appraisal), data-driven decision support (by analyzing historical data and real-time data, providing scientific decision-making basis, reducing human bias, and improving the effectiveness of performance management), and improving appraisal efficiency (optimizing the appraisal process, reducing human intervention, and improving the overall work efficiency of the system, so that management can focus on higher-level strategic decisions).

[0014] The present invention achieves the above technical objectives through the following technical means.

[0015] A dynamic performance appraisal adjustment method based on machine learning includes the following process:

[0016] Step 1: The data acquisition and processing module obtains raw data from internal and external data sources;

[0017] Step 2: The data acquisition and processing module cleans and normalizes the acquired data, and then loads the processed data into a data warehouse or analysis platform for subsequent analysis and modeling;

[0018] Step 3: The feature extraction and model building module extracts the corresponding features from the raw data processed in step 2 through machine learning algorithms and builds a performance appraisal prediction model to support data-driven decision-making;

[0019] Step 4: The big data analysis and real-time adjustment module relies on the prediction model built in step 3 to conduct in-depth analysis of the large-scale data processed in step 2 and dynamically adjust the performance evaluation standards;

[0020] Step 5: The big data analysis and real-time adjustment module further dynamically optimizes and updates the prediction model based on the analysis results of step 4 to adapt to environmental changes and new data input, so that the prediction model has a better evaluation effect;

[0021] Step 6: After the real-time adjustment is completed, the big data analysis and real-time adjustment modules make automated decisions to further improve the system response speed and efficiency;

[0022] Step 7: The system monitoring and maintenance module tracks and evaluates the system's performance indicators in real time to ensure the system's stability, reliability, and adaptability.

[0023] Furthermore, the specific process of step 3 is as follows:

[0024] Step 3.1: Feature extraction;

[0025] Step 3.1.1: Adaptive dynamic feature selection mechanism: filter irrelevant or redundant features to enhance the adaptability of the prediction model; in terms of feature selection, an enhanced version of Lasso regression combined with random forest algorithm is used;

[0026] Step 3.1.2: Feature engineering optimization: On the selected features, the following optimization methods are used: numerical feature conversion, category feature encoding, and time series feature extraction;

[0027] Step 3.2: Prediction model construction;

[0028] Step 3.2.1: Algorithm selection and adaptability enhancement: After feature selection, ARIMA and LSTM models are selected for short-term and long-term time series prediction respectively;

[0029] Step 3.2.2: Weighted combination of hybrid models: Based on the ARIMA and LSTM models, a weighted hybrid model is introduced to perform a weighted combination of the prediction results of the two to improve the accuracy of the prediction. The specific method is: based on the prediction errors of historical and current data, the weights of the prediction results of ARIMA and LSTM are dynamically adjusted to form a combined prediction value. The dynamic weight adjustment is optimized according to the error of each model to adapt to different data distributions and change trends;

[0030] Step 3.2.3: Prediction model training;

[0031] Step 3.3: Prediction model evaluation and optimization;

[0032] Step 3.4: Prediction model deployment and application, prediction model monitoring and retraining.

[0033] Furthermore, the specific process of step 3.1.1 is as follows:

[0034] First, set the feature space and target variable;

[0035] Then, the improved Lasso regression is used for preliminary screening to obtain a set of candidate features, ensuring high correlation and eliminating redundant features. The optimization goal of the improved Lasso regression is:

[0036]

[0037] In the formula, λ is the regularization parameter used to control the strength of feature selection; i represents the index of the data point; m represents the total number of samples in the data set, the number of data points; y i represents the observed value of the response variable at the ith data point; x ijrepresents the jth eigenvalue of the i-th data point; β j represents the regression coefficient of the jth feature; j represents the index of the feature; n represents the total number of features;

[0038] Then, based on the features screened by Lasso regression, the feature importance is sorted in the random forest model for the candidate feature set, the feature set is further optimized, and the threshold T is set. importance To determine the optimal feature set F * :

[0039] F * ={x j |importance(x j )>T importance}

[0040] In the formula, x j represents the jth feature; importance(x j ) represents the feature x j The importance value of

[0041] In practical applications, the importance of features is regularly re-evaluated according to data fluctuations. * Make dynamic adjustments.

[0042] Furthermore, the specific process of step 4 is as follows:

[0043] Step 4.1: Data processing and integration: Combine the collected data with the constructed prediction model to conduct in-depth analysis of the performance data and make dynamic adjustments based on the analysis results;

[0044] Step 4.2: Time series forecasting and trend analysis;

[0045] Step 4.2.1: Use the Autoregressive Integrated Moving Average Model (ARIMA) model and Long Short-Term Memory (LSTM) model trained in Step 3 to predict the performance trend of the data stream, so as to identify the potential trend of performance changes in advance;

[0046] Step 4.2.2: Hybrid model forecasting: Use the weighted forecast formula of ARIMA and LSTM to calculate the hybrid forecast value Dynamically adjust the prediction weights based on historical and current data. The weights are adjusted based on the prediction error of each model to ensure the accuracy of the model predictions:

[0047]

[0048] Among them, wARIMA is the ARIMA weight, w LSTM is the weight of LSTM; represents the predicted value of the ARIMA model at time t; Represents the predicted value of the LSTM model at time t; Represents the final forecast value of the hybrid model, which is the result of a weighted combination of the ARIMA and LSTM models.

[0049] Step 4.2.3: Anomaly detection and alerting: Combine the deviation between the predicted value and the current actual data to automatically detect abnormal performance changes. If the deviation exceeds the set threshold, the alert mechanism is triggered so that management can intervene;

[0050] Step 4.3: Dynamic threshold adjustment and model adaptive update;

[0051] Step 4.3.1: Implement threshold adjustment mechanism: Introduce exponential smoothing algorithm to adaptively adjust the dynamic threshold of performance standards, so that performance evaluation can timely reflect the fluctuations of employees and the market, and improve the sensitivity and accuracy of the performance appraisal prediction model. The real-time threshold calculation formula is:

[0052] T t =α×KPI t +(1-α)×T t-1

[0053] Among them, α is a smoothing factor, which is applicable to the performance appraisal needs of different departments and positions; KPI t represents the key performance indicator at time t; T t They represent the dynamic threshold at time t. The threshold will be adjusted in real time according to the needs of performance appraisal to adapt to changes in employees and the market. t-1 Indicates the dynamic threshold at time t-1, that is, the threshold at the previous time point, which is used to smooth and continue the change trend of the threshold.

[0054] Step 4.3.2: Model adaptive update: Based on the actual performance feedback from step 4.2.3 and the data distribution changes feedback from step 4.2.1, the prediction model is retrained and optimized to ensure that the prediction model can maintain accuracy and sensitivity; the adaptive update cycle can be dynamically adjusted according to the speed of data changes or market demand.

[0055] Furthermore, the specific process of step 5 is as follows:

[0056] Step 5.1: Model update and retraining: When the data or environment changes significantly, step 5 will retrain or adjust the parameters of the prediction model to improve its adaptability based on the data distribution changes or model deviations detected in step 4;

[0057] Step 5.2: Dynamic adjustment of thresholds and standards: Use the analysis results of step 4 to dynamically adjust the assessment thresholds or standards so that the performance assessment of the prediction model is more in line with the fluctuations of real-time data;

[0058] Step 5.3: Automated decision-making: Based on the analysis in step 4, provide optimization parameters for automated decision-making and feedback so that the system (the entire predictive model and its associated automated decision-making module and feedback mechanism) can generate personalized performance feedback and management recommendations more accurately.

[0059] A dynamic performance appraisal adjustment system based on machine learning for implementing the above-mentioned dynamic performance appraisal adjustment method based on machine learning, comprising: a data acquisition and processing module, a feature extraction and model building module, a big data analysis and real-time adjustment module, and a system monitoring and maintenance module;

[0060] The data acquisition and processing module is used to obtain raw data from internal and external data sources, and perform preliminary processing and storage;

[0061] The feature extraction and model building module identifies and extracts meaningful features from the original data, and selects the corresponding machine learning algorithm to train the model and build a prediction model based on the features;

[0062] The big data analysis and real-time adjustment module conducts in-depth analysis of real-time input data based on the prediction model to help optimize decision-making, and dynamically adjusts the prediction model or strategy based on the analysis results to adapt to the changing environment;

[0063] The system monitoring and maintenance module ensures that the dynamic performance appraisal adjustment system operates efficiently and stably in a complex business environment through real-time performance monitoring, data quality management, automated fault recovery and feedback optimization mechanisms.

[0064] Furthermore, the system monitoring and maintenance module includes a model monitoring submodule, a data monitoring submodule, a system operation monitoring submodule, and a feedback and optimization submodule;

[0065] The model monitoring submodule is used to implement prediction model performance monitoring, monitor the performance of the prediction model in the production environment, evaluate key indicators, and determine whether the prediction model meets expectations. It is also used for anomaly detection. When there is a significant deviation between the prediction results of the prediction model and the actual results, an alarm is triggered. It is also used to trigger model retraining. When a change in data distribution or a decrease in prediction model performance is detected, the model retraining mechanism is automatically triggered and linked with the prediction model construction module.

[0066] The data monitoring submodule is used to monitor the quality of the data provided by the data acquisition and processing module, and detect changes in data distribution. When the distribution of real-time data deviates significantly from the distribution of historical data, the data monitoring submodule records and marks it, and triggers the update of the prediction model when necessary;

[0067] The system operation monitoring submodule is used to monitor system performance indicators, including processing delay, system response time, and throughput; it is also used to achieve automatic fault recovery;

[0068] The feedback and optimization submodule integrates the feedback from mobile phone users or management on the performance appraisal adjustment system, and further improves the applicability of the system through data analysis and optimization suggestions. It is also used to analyze the deviation between automated decisions and actual results, and adjust the appraisal standards or prediction model parameters accordingly to continuously optimize system decisions.

[0069] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the dynamic performance appraisal adjustment method based on machine learning as described in claim 1.

[0070] The present invention has the following beneficial effects:

[0071] Automated performance standard adjustment based on machine learning: Using machine learning algorithms, the system can analyze and automatically adjust performance standards in real time according to project characteristics and market dynamics, thereby ensuring that performance appraisals are consistent with actual conditions and improving the adaptability of appraisals.

[0072] Dynamically optimized performance appraisal process: By dynamically optimizing the performance appraisal process, the system can flexibly respond to different projects and environmental changes, ensuring that the appraisal process is not only accurate but also able to quickly respond to new challenges and demands.

[0073] Data-driven decision support: Strengthen data-driven decision support and continuously optimize performance management strategies through continuous learning and real-time data analysis. This approach enables management to make more informed decisions based on actual data and improve overall project management efficiency.

[0074] The use of the dynamic performance appraisal adjustment system of the present invention will significantly improve the scientificity, accuracy and efficiency of enterprise performance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 This is a flow chart of the dynamic performance appraisal adjustment method based on machine learning described in the present invention. DETAILED DESCRIPTION

[0076] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited thereto.

[0077] The dynamic performance appraisal adjustment system based on machine learning described in the present invention includes a data acquisition and processing module, a feature extraction and model building module, a big data analysis and real-time adjustment module, and a system monitoring and maintenance module.

[0078] The data acquisition and processing module is used to acquire, clean and integrate data from internal and external data sources to ensure the accuracy and availability of the data. The data acquisition and processing module includes functions such as data preprocessing, data conversion, data integration and data storage. Its goal is to provide high-quality basic data for subsequent analysis and ensure data consistency and integrity.

[0079] Among them, internal data sources include internal company data (the company's own human resources information, including employee files, attendance records, and performance evaluations), customer feedback data (including satisfaction surveys and complaint records from customers), project documents (including contracts, meeting minutes, and change orders), and information system data (including data stored in ERP and CRM systems).

[0080] External data sources include industry reports (industry analysis and trend reports published by market research organizations), government statistics (economic, demographic, industry and other related statistics released by national and local governments), academic research (research papers and survey results from universities and research institutions), market surveys (market surveys and consumer behavior analysis conducted by third-party companies), social media and online reviews (user feedback and comments on social platforms and websites), news and media (industry news, professional magazines and website reports), and professional conferences and exhibitions (data and information collected at industry activities).

[0081] The feature extraction and model building module identifies and extracts meaningful features from the raw data, and selects appropriate machine learning algorithms for model training based on the features to improve the performance of the model. The feature extraction and model building module includes functions such as feature selection, feature engineering, model training and evaluation. Its goal is to build an efficient prediction model to support the decision-making process and improve the accuracy of project management.

[0082] The big data analysis and real-time adjustment module extracts valuable information from massive data to help optimize decision-making, and dynamically adjusts models or strategies based on analysis results to adapt to changing environments. The big data analysis and real-time adjustment module includes functions such as data processing, real-time analysis, dynamic model updates, and automated decision-making. Its goal is to improve the system's responsiveness and decision-making efficiency, so that project management can adapt to environmental changes and data fluctuations in a timely manner.

[0083] The system monitoring and maintenance module ensures that the dynamic performance appraisal system operates efficiently and stably in a complex business environment through real-time performance monitoring, data quality management, automated fault recovery, and feedback optimization mechanisms. It has flexible adaptability and feedback-driven optimization capabilities, can quickly respond to data fluctuations and demand changes, and provide guarantees for the long-term reliability and accuracy of the system.

[0084] Reference Figure 1 The dynamic performance appraisal adjustment method based on machine learning using the dynamic performance appraisal adjustment system based on machine learning includes the following process:

[0085] Step 1: The data acquisition and processing module obtains raw data from internal and external data sources;

[0086] By writing scripts or using data collection tools, data is automatically obtained from internal and external data sources on a regular basis. Common technical means include: API calls, web crawlers, and database connections; API calls use the API interfaces provided by various data sources to extract data regularly to ensure timeliness of data updates; web crawlers use web crawler technology to capture required information from websites, which is particularly suitable for public data that does not provide APIs; database connections are to establish direct connections with the company's internal databases, regularly extract required data, and maintain data consistency.

[0087] Step 2: The data acquisition and processing module cleans and normalizes the acquired data, and then loads the processed data into the data warehouse or analysis platform for subsequent analysis and modeling; the details are as follows:

[0088] Data cleaning:

[0089] (1) Data preprocessing:

[0090] Deduplication: Identify and delete duplicate data to ensure the uniqueness of the data set;

[0091] Filling: Use mean, median or other strategies to fill missing values ​​to ensure data integrity;

[0092] Outlier handling: Detect and handle outliers, possibly by replacing, deleting, or adjusting to reduce their impact;

[0093] (2) Data conversion:

[0094] Data format conversion and data normalization.

[0095] Data normalization: converting data of different ranges or scales to a uniform range, usually [0,1] or [-1,1]. Common methods include:

[0096] (1) Feature Standardization

[0097] Min-max normalization: Transform the data using the following formula:

[0098]

[0099] Among them, x is the original data, X is the data set, and x′ represents the data after minimum-maximum normalization;

[0100] Z-score standardization: The data is made to conform to a standard normal distribution by subtracting the mean and dividing by the standard deviation:

[0101]

[0102] Among them, μ is the mean, σ is the standard deviation, and z is the data after Z-score standardization;

[0103] Decimal scaling method: divide the data by 10 to the power of n to make the data within a specific range;

[0104] (2) Data integration and conversion

[0105] Merge data sources: Integrate data from different sources to form a complete data set;

[0106] Feature transformation: Extract more meaningful information by combining and transforming features.

[0107] Step 3: The feature extraction and model building module extracts meaningful features from the raw data processed in step 2 through machine learning algorithms, and builds an efficient performance appraisal prediction model (hereinafter referred to as the prediction model) to support data-driven decision-making; the specific process is as follows:

[0108] Step 3.1: Feature extraction: Identify and extract meaningful information from raw data to improve the performance of the prediction model;

[0109] Step 3.1.1: Adaptive dynamic feature selection mechanism: In order to enhance the adaptability of the prediction model, irrelevant or redundant features are filtered out; in terms of feature selection, an enhanced version of Lasso regression combined with random forest algorithm is used. The specific process is as follows:

[0110] First, set the feature space X = {x1, x2, ..., x n} and target variable Y;

[0111] Then, the improved Lasso regression is used for preliminary screening to obtain the candidate feature set S, ensuring high correlation and eliminating redundant features. The optimization goal of the improved Lasso regression is:

[0112]

[0113] In the formula, λ is the regularization parameter used to control the strength of feature selection; i represents the index of the data point; m represents the total number of samples in the data set, the number of data points; y i represents the observed value of the response variable of the ith data point, which represents the actual result that the prediction model is expected to predict; x ij represents the jth eigenvalue of the i-th data point, which represents the input feature variable used to predict the result; β j represents the regression coefficient of the jth feature, indicating the influence or weight of the feature on the prediction result; j represents the index of the feature; n represents the total number of features, the number of features contained in the data set;

[0114] Then, based on the features screened by Lasso regression, the feature importance is sorted in the random forest model for the candidate feature set S, the feature set is further optimized, and the threshold T is set. importance To determine the optimal feature set F * :

[0115] F * ={x j |importance(x j )>T importance}

[0116] In the formula, x j represents the jth feature; importance(x j ) represents the feature x j The importance value of feature x is usually calculated by a certain model (such as random forest) and reflects the j Relative importance in the forecast;

[0117] In practical applications, the importance of features is regularly re-evaluated according to data fluctuations. * Make dynamic adjustments.

[0118] Step 3.1.2: Feature engineering optimization: On the selected features, the following optimization methods are used:

[0119] Numerical feature transformation: Apply logarithmic transformation, square root transformation and other methods to process skewed distribution features;

[0120] Categorical feature encoding: Use one-hot encoding, label encoding, or target encoding to convert categorical features into numerical form;

[0121] Time series feature extraction: Extract useful information from timestamps, such as year, month, day, week, etc.

[0122] Step 3.2: Prediction model construction:

[0123] Step 3.2.1: Algorithm selection and adaptability enhancement: After feature selection is completed, ARIMA and LSTM models are selected for short-term and long-term time series prediction respectively:

[0124] ARIMA model: Suitable for short-term time series forecasting, mainly used to capture seasonal changes and trend fluctuations in data.

[0125] LSTM model: used to process the dependencies of long time series and better predict long-term trends. By combining short-term and long-term models, it improves the adaptability to data of different time scales.

[0126] Step 3.2.2: Weighted combination of mixed models:

[0127] Based on the ARIMA and LSTM models, a weighted hybrid model is introduced to weight the prediction results of the two models, thereby improving the accuracy of the prediction. The specific operations are as follows:

[0128] Based on the prediction errors of historical and current data, the weights of ARIMA and LSTM prediction results are dynamically adjusted to form a combined prediction value. Dynamic weight adjustment can be optimized according to the error of each model to adapt to different data distributions and changing trends.

[0129] Step 3.2.3: Prediction model training process:

[0130] Dataset division: The data set is divided into training and test sets, and the ARIMA and LSTM models are trained, and cross-validation and hyperparameter optimization are used to improve model performance. At the same time, the weighted combination prediction results of the two are combined, and the model is evaluated using indicators to ensure its accuracy at different time scales.

[0131] Cross-validation and hyperparameter tuning: Use cross-validation to evaluate the generalization ability of the model, and use methods such as grid search or Bayesian optimization to adjust hyperparameters to improve model performance.

[0132] Step 3.3: Prediction model evaluation and optimization:

[0133] Step 3.3.1: Evaluation indicators: Select appropriate indicators according to the model task, such as accuracy, mean square error (MSE), F1-score, etc., to ensure that the model performance meets the expected requirements;

[0134] Step 3.3.2: Overfitting and underfitting processing:

[0135] Reduce overfitting through regularization (L2 regularization), dropout, data enhancement, etc.

[0136] For underfitting, increase feature complexity or use a deeper neural network structure;

[0137] Adaptive model evaluation: Use real-time data to verify the actual effect of the model, and regularly evaluate and adjust the model according to the distribution changes of new data to ensure its stability and effectiveness.

[0138] Step 3.4: Prediction model deployment and application:

[0139] Step 3.4.1: Model saving: Use Pickle, ONNX or other formats to save the trained model as a reusable file for subsequent calls;

[0140] Step 3.4.2: Real-time prediction API interface to ensure that the system can quickly access real-time data and output results;

[0141] Step 3.4.3: Prediction model monitoring and retraining:

[0142] Monitor model performance: Regularly monitor the performance of the model in the production environment to ensure that its predictive ability does not decrease with data fluctuations;

[0143] Retraining mechanism: When the data distribution changes significantly, the retraining process is triggered to maintain the adaptability of the model.

[0144] Step 4: The big data analysis and real-time adjustment module relies on the prediction model built in step 3 to conduct in-depth analysis of the large-scale data processed in step 2 to identify trends, patterns and insights, and dynamically adjust the performance evaluation standards; the specific process is as follows:

[0145] Step 4.1: Data processing and integration: The big data analysis and real-time adjustment module combines the collected data with the constructed prediction model to conduct in-depth analysis of the performance data and make dynamic adjustments based on the analysis results;

[0146] Step 4.1.1: Distributed data processing: Use distributed computing frameworks such as Apache Spark or Hadoop to process large-scale data sets to improve processing efficiency;

[0147] Step 4.1.2: Real-time data stream processing: Pre-process the real-time data stream through Kafka or Flink to ensure that the newly input data can enter the system in time;

[0148] Step 4.1.3: Feature engineering application: Apply the feature engineering results in step 3 to real-time data to ensure data format consistency and availability, and provide a good foundation for model analysis.

[0149] Step 4.2: Time Series Forecasting and Trend Analysis:

[0150] Step 4.2.1: Use the Autoregressive Integrated Moving Average Model (ARIMA) model and Long Short-Term Memory (LSTM) model trained in Step 3 to predict the performance trend of the data stream, so as to identify the potential trend of performance changes in advance;

[0151] Step 4.2.2: Hybrid model forecasting: Use the weighted forecast formula of ARIMA and LSTM to calculate the hybrid forecast value Dynamically adjust the prediction weights based on historical and current data. The weights are adjusted based on the prediction error of each model to ensure the accuracy of the model predictions:

[0152]

[0153] Among them, w ARIMA is the ARIMA weight, w LSTM is the weight of LSTM; represents the predicted value of the ARIMA model at time t; Represents the predicted value of the LSTM model at time t; Represents the final forecast value of the hybrid model, which is the result of a weighted combination of the ARIMA and LSTM models.

[0154] Step 4.2.3: Anomaly detection and reminder: Combine the deviation between the predicted value and the current actual data to automatically detect abnormal performance changes. If the deviation exceeds the set threshold, the reminder mechanism is triggered so that management can intervene.

[0155] Step 4.3: Dynamic threshold adjustment and model adaptive update:

[0156] Step 4.3.1: Implement threshold adjustment mechanism: Introduce exponential smoothing algorithm to adaptively adjust the dynamic threshold of performance standards, so that performance evaluation can timely reflect the fluctuations of employees and the market, and improve the sensitivity and accuracy of the performance appraisal prediction model. The real-time threshold calculation formula is:

[0157] T t =α×KPI t +(1-α)×T t-1

[0158] Among them, α is a smoothing factor, which is applicable to the performance appraisal needs of different departments and positions; KPI t represents the key performance indicator at time t; T t They represent the dynamic threshold at time t. The threshold will be adjusted in real time according to the needs of performance appraisal to adapt to changes in employees and the market. t-1Indicates the dynamic threshold at time t-1, that is, the threshold at the previous time point, which is used to smooth and continue the change trend of the threshold.

[0159] Step 4.3.2: Model adaptive update: Based on the actual performance feedback from step 4.2.3 and the data distribution changes feedback from step 4.2.1, the prediction model is retrained and optimized to ensure that the prediction model can maintain accuracy and sensitivity; the adaptive update cycle can be dynamically adjusted according to the speed of data changes or market demand.

[0160] Step 5: The big data analysis and real-time adjustment module further dynamically optimizes and updates the prediction model based on the analysis results of step 4 to adapt to environmental changes and new data input, so that the prediction model has a better evaluation effect; the specific process is as follows:

[0161] Step 5.1: Model update and retraining: When the data or environment changes significantly, step 5 will retrain or adjust the parameters of the prediction model to improve its adaptability based on the data distribution changes or model deviations detected in step 4;

[0162] Step 5.2: Dynamic adjustment of thresholds and standards: Use the analysis results of step 4 to dynamically adjust the assessment thresholds or standards so that the performance assessment of the prediction model is more in line with the fluctuations of real-time data;

[0163] Step 5.3: Automated decision-making: Based on the analysis in step 4, provide optimization parameters for automated decision-making and feedback so that the system (the entire predictive model and its associated automated decision-making module and feedback mechanism) can generate personalized performance feedback and management recommendations more accurately.

[0164] Step 6: After the real-time adjustment is completed, the big data analysis and real-time adjustment modules make automated decisions to further improve the system's response speed and efficiency;

[0165] Step 6.1: Rules engine: Automates decision making based on predefined rules, such as early warning mechanisms based on key performance indicators (KPIs);

[0166] Step 6.2: Intelligent recommendation system: Provide personalized recommendations by analyzing user behavior and preferences to improve user experience and satisfaction;

[0167] Step 7: The system monitoring and maintenance module is responsible for real-time tracking and evaluation of various performance indicators of the system to ensure the stability, reliability and adaptability of the system. Its implementation methods include the following aspects:

[0168] Step 7.1: Model monitoring submodule:

[0169] Step 7.1.1: Model performance monitoring: The model monitoring submodule monitors the performance of the prediction model in the production environment, evaluates key indicators (such as accuracy, F1 score, MSE, etc.), and determines whether the prediction model meets expectations;

[0170] Step 7.1.2: Anomaly detection: When the prediction results of the prediction model deviate significantly from the actual results, an alarm is triggered; statistical methods or machine learning methods can be used for anomaly detection;

[0171] Step 7.1.3: Model retraining trigger: Once the model monitoring submodule detects a change in data distribution or a decline in the performance of the prediction model, it automatically triggers the model retraining mechanism and works in conjunction with the prediction model building module.

[0172] Step 7.2: Data monitoring submodule:

[0173] Step 7.2.1: Data quality monitoring: The data monitoring submodule ensures the integrity and consistency of the data provided by the data acquisition and processing module, monitors whether there are problems such as data loss, duplication, and outliers, and ensures the quality of data processing;

[0174] Step 7.2.2: Data distribution change detection: When the distribution of real-time data deviates significantly from the distribution of historical data, the data monitoring submodule records and marks it, and triggers a model update when necessary.

[0175] Step 7.3: System operation monitoring submodule:

[0176] Step 7.3.1: Monitor system performance indicators: including processing delay, system response time, throughput, etc., to ensure that the system runs stably under high load;

[0177] Step 7.3.2: Automatic fault recovery: When the system detects an error crash, it automatically starts the preset recovery mechanism to ensure the continuous high availability of the system.

[0178] Step 7.4: Feedback and Optimization Submodule:

[0179] Step 7.4.1: User feedback integration: The feedback and optimization submodule integrates the feedback from mobile phone users or management on the performance appraisal adjustment system, and further improves the applicability of the system through data analysis and optimization suggestions;

[0180] Step 7.4.2: Decision feedback loop: The feedback and optimization submodule analyzes the deviation between the automated decision in step 4 and the actual result, and adjusts the assessment criteria or model parameters accordingly to continuously optimize the system decision.

[0181] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the above-mentioned dynamic performance appraisal adjustment method based on machine learning are implemented.

[0182] The embodiments are preferred implementations of the present invention, but the present invention is not limited to the above-mentioned implementations. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essential content of the present invention belong to the protection scope of the present invention.

Claims

1. A dynamic performance appraisal adjustment method based on machine learning, characterized in that: The process includes the following: Step 1: The data acquisition and processing module obtains raw data from internal and external data sources; Step 2: The data acquisition and processing module cleans and normalizes the acquired data, and then loads the processed data into a data warehouse or analysis platform for subsequent analysis and modeling; Step 3: The feature extraction and model building module extracts the corresponding features from the raw data processed in step 2 through machine learning algorithms and builds a performance appraisal prediction model to support data-driven decision-making; Step 4: The big data analysis and real-time adjustment module relies on the prediction model built in step 3 to conduct in-depth analysis of the large-scale data processed in step 2 and dynamically adjust the performance evaluation standards; Step 5: The big data analysis and real-time adjustment module further dynamically optimizes and updates the prediction model based on the analysis results of step 4 to adapt to environmental changes and new data input, so that the prediction model has a better evaluation effect; Step 6: After the real-time adjustment is completed, the big data analysis and real-time adjustment modules make automated decisions to further improve the system response speed and efficiency; Step 7: The system monitoring and maintenance module tracks and evaluates the system's performance indicators in real time to ensure the system's stability, reliability, and adaptability.

2. The dynamic performance appraisal adjustment method based on machine learning according to claim 1 is characterized in that: The specific process of step 3 is as follows: Step 3.1: Feature extraction; Step 3.1.1: Adaptive dynamic feature selection mechanism: filter irrelevant or redundant features to enhance the adaptability of the prediction model; In terms of feature selection, an enhanced version of Lasso regression combined with random forest algorithm is used; Step 3.1.2: Feature engineering optimization: On the selected features, the following optimization methods are used: numerical feature conversion, category feature encoding, and time series feature extraction; Step 3.2: Prediction model construction; Step 3.2.1: Algorithm selection and adaptability enhancement: After feature selection, ARIMA and LSTM models are selected for short-term and long-term time series prediction respectively; Step 3.2.2: Weighted combination of hybrid models: Based on the ARIMA and LSTM models, a weighted hybrid model is introduced to perform a weighted combination of the prediction results of the two to improve the accuracy of the prediction. The specific method is: based on the prediction errors of historical and current data, the weights of the prediction results of ARIMA and LSTM are dynamically adjusted to form a combined prediction value. The dynamic weight adjustment is optimized according to the error of each model to adapt to different data distributions and change trends; Step 3.2.3: Prediction model training; Step 3.3: Prediction model evaluation and optimization; Step 3.4: Prediction model deployment and application, prediction model monitoring and retraining.

3. The dynamic performance appraisal adjustment method based on machine learning according to claim 2 is characterized in that: The specific process of step 3.1.1 is as follows: First, set the feature space and target variable; Then, the improved Lasso regression is used for preliminary screening to obtain a set of candidate features, ensuring high correlation and eliminating redundant features. The optimization goal of the improved Lasso regression is: In the formula, λ is the regularization parameter used to control the strength of feature selection; i represents the index of the data point; m represents the total number of samples in the data set, the number of data points; y i represents the observed value of the response variable at the ith data point; x ij represents the jth eigenvalue of the i-th data point; β j represents the regression coefficient of the jth feature; j represents the index of the feature; n represents the total number of features; Then, based on the features screened by Lasso regression, the importance of features is sorted in the random forest model for the candidate feature set, and the feature set is further optimized, and the threshold T is set. importance To determine the optimal feature set F * : F * ={x j |impor tan ce(x j )>T impor tan ce } In the formula, x j represents the jth feature; impact(x j ) represents the feature x j The importance value of In practical applications, the importance of features is regularly re-evaluated according to data fluctuations. * Make dynamic adjustments.

4. The dynamic performance appraisal adjustment method based on machine learning according to claim 1 is characterized in that: The specific process of step 4 is as follows: Step 4.1: Data processing and integration: Combine the collected data with the constructed prediction model to conduct in-depth analysis of the performance data and make dynamic adjustments based on the analysis results; Step 4.2: Time series forecasting and trend analysis; Step 4.2.1: Use ARIMA and LSTM models to predict performance trends of data streams in order to identify potential trends of performance changes in advance; Step 4.2.2: Hybrid Model Forecasting: Use the weighted forecast formula of ARIMA and LSTM to calculate the hybrid forecast value Dynamically adjust the prediction weights based on historical and current data. The weights are adjusted based on the prediction error of each model to ensure the accuracy of the model predictions: Among them, w ARIMA is the ARIMA weight, w LSTM is the weight of LSTM; represents the predicted value of the ARIMA model at time t; Represents the predicted value of the LSTM model at time t; Represents the final forecast value of the hybrid model, which is the result of the weighted combination of the ARIMA and LSTM models; Step 4.2.3: Anomaly detection and alerting: Combine the deviation between the predicted value and the current actual data to automatically detect abnormal performance changes. If the deviation exceeds the set threshold, the alert mechanism is triggered so that management can intervene; Step 4.3: Dynamic threshold adjustment and model adaptive update; Step 4.3.1: Implement threshold adjustment mechanism: Introduce exponential smoothing algorithm to adaptively adjust the dynamic threshold of performance standards, so that performance evaluation can timely reflect the fluctuations of employees and the market, and improve the sensitivity and accuracy of the performance appraisal prediction model. The real-time threshold calculation formula is: T t =α×KPI t +(1-a)×T t-1 Among them, α is a smoothing factor, which is applicable to the performance appraisal needs of different departments and positions; KPI t represents the key performance indicator at time t; T t Respectively represent the dynamic threshold at time t; T t-1 represents the dynamic threshold at time t-1; Step 4.3.2: Model adaptive update: Retrain and optimize the prediction model based on the actual performance feedback from step 4.2.3 and the data distribution changes feedback from step 4.2.

1.

5. A machine learning-based dynamic performance appraisal adjustment system for implementing the machine learning-based dynamic performance appraisal adjustment method described in step 1, characterized in that: include: Data collection and processing module, feature extraction and model building module, big data analysis and real-time adjustment module, system monitoring and maintenance module; The data acquisition and processing module is used to obtain raw data from internal and external data sources, and perform preliminary processing and storage; The feature extraction and model building module identifies and extracts meaningful features from the original data, and selects the corresponding machine learning algorithm to train the model and build a prediction model based on the features; The big data analysis and real-time adjustment module conducts in-depth analysis of real-time input data based on the prediction model to help optimize decision-making, and dynamically adjusts the prediction model or strategy based on the analysis results to adapt to the changing environment; The system monitoring and maintenance module ensures that the dynamic performance appraisal adjustment system operates efficiently and stably in a complex business environment through real-time performance monitoring, data quality management, automated fault recovery and feedback optimization mechanisms.

6. The dynamic performance evaluation adjustment system based on machine learning according to claim 5 is characterized in that: The system monitoring and maintenance module includes a model monitoring submodule, a data monitoring submodule, a system operation monitoring submodule, and a feedback and optimization submodule; The model monitoring submodule is used to implement prediction model performance monitoring, monitor the performance of the prediction model in the production environment, evaluate key indicators, and determine whether the prediction model meets expectations; It is also used for anomaly detection. When the prediction results of the prediction model deviate significantly from the actual results, an alarm is triggered. It is also used to trigger model retraining. When changes in data distribution or a decline in the performance of the prediction model are detected, the model retraining mechanism is automatically triggered and linked with the prediction model construction module. The data monitoring submodule is used to monitor the quality of the data provided by the data acquisition and processing module, and detect changes in data distribution. When the distribution of real-time data deviates significantly from the distribution of historical data, the data monitoring submodule records and marks it, and triggers the update of the prediction model when necessary; The system operation monitoring submodule is used to monitor system performance indicators, including processing delay, system response time, and throughput; It is also used to achieve automatic fault recovery; The feedback and optimization submodule integrates the feedback from mobile phone users or management on the performance appraisal adjustment system, and further improves the applicability of the system through data analysis and optimization suggestions. It is also used to analyze the deviation between automated decisions and actual results, and adjust the appraisal standards or prediction model parameters accordingly to continuously optimize system decisions.

7. A computer-readable storage medium, characterized in that: A computer program is stored thereon, which, when executed by a processor, implements the steps of the dynamic performance appraisal adjustment method based on machine learning as described in claim 1.

Images