A method and system for processing store sales data

Abstract

This invention relates to the field of image data processing technology, and more particularly to a method and system for processing store sales data. The method includes: acquiring reference samples and actual sales figures for each store, as well as samples to be predicted for which store sales figures need to be predicted; training a random forest algorithm and using the trained model for prediction; designating any prediction tree in the trained model as the target tree and determining the accuracy of the target tree; determining the weight factors for the samples to be predicted when using the target tree for prediction; and determining the final predicted sales figures for the samples to be predicted, thereby achieving store sales data processing. This invention dynamically allocates weight factors for each tree by comprehensively considering multiple dimensions such as the training sample coverage, prediction error of unselected samples, and sample path homogeneity of a single decision tree, avoiding interference from low-quality predictions on the final result and improving the accuracy and reliability of store sales prediction.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, and in particular to a method and system for processing store sales data. Background Technology

[0002] In the fiercely competitive landscape of modern retail, consumer behavior is becoming increasingly complex and volatile, making store sales forecasting a core competitive advantage for businesses seeking to gain a competitive edge. Traditional forecasting methods rely heavily on human experience and manual data analysis, which struggles to cope with the massive amounts of data, complex information, and ever-changing market environment. With the deep application of big data and machine learning technologies, building efficient and accurate intelligent forecasting models has become a crucial path for retail enterprises to achieve dynamic inventory management, optimize marketing strategies, and improve operational efficiency. The Random Forest Algorithm (RF), with its powerful data processing capabilities, can automatically filter core features from multi-dimensional data (such as customer traffic, store area, promotional budgets, and historical sales data), and reduce model variance by integrating the prediction results of multiple decision trees, effectively improving prediction stability. Through bootstrap sampling and random feature selection mechanisms, the Random Forest algorithm ensures that each decision tree focuses on different data distributions and feature combinations during training, thereby enhancing the model's adaptability to complex scenarios.

[0003] Traditional random forest models use an equal-weighted averaging strategy to integrate the prediction results of individual decision trees. This means simply taking the arithmetic average of the prediction results from each decision tree as the final output. However, each decision tree uses different reference samples and features for training, leading to differences in prediction accuracy. For example, some decision trees may perform poorly in specific scenarios (such as major promotional events or seasonal fluctuations) due to insufficient training sample coverage or biased feature selection. Simply averaging these results will dilute the contribution of high-accuracy models. Furthermore, different decision trees have different applicability to the same sample to be predicted. For instance, a decision tree that frequently encounters data patterns similar to the sample to be predicted during training should have higher prediction reliability than other decision trees that do not cover this pattern. Directly averaging these results will cause the prediction to deviate from the actual value, thus affecting the prediction of store sales. Summary of the Invention

[0004] To address the issues that traditional random forest models use an equal-weighted averaging strategy to integrate the prediction results of individual decision trees, but each decision tree uses different reference samples and feature selections for training, leading to differences in prediction accuracy; and that different decision trees have varying degrees of applicability to the same sample to be predicted, direct averaging can cause the prediction results to deviate from the actual values, thus affecting the prediction of store sales, this invention provides a method and system for processing store sales data.

[0005] In a first aspect, the present invention provides a method for processing store sales data, employing the following technical solution:

[0006] A method for processing store sales data includes: obtaining reference samples and actual sales of the reference samples for each store, and samples to be predicted for which store sales need to be predicted, wherein the reference samples and samples to be predicted contain multiple relevant operational features; training a random forest algorithm based on the reference samples to obtain a store sales prediction model; and using the store sales prediction model to predict the samples to be predicted, including: designating any prediction tree in the store sales prediction model as the target tree, and based on the number of duplicate reference samples in the reference samples selected by the target tree, the number of reference samples selected by the target tree, the number of reference samples not selected by the target tree, and the prediction of each reference sample in the unselected reference samples of the target tree using... The difference between the predicted sales revenue obtained from the target tree and the actual sales revenue determines the accuracy of the target tree. Based on the accuracy, the values ​​of the relevant operational features corresponding to all split points of the target tree for the sample to be predicted, the values ​​of the relevant operational features corresponding to all split points of the target tree, and the number of reference samples with the same path as the sample to be predicted during the training process of the target tree, the weight factor for the sample to be predicted when using the target tree for prediction is determined. Based on the weight factor, the predicted sales revenue obtained by the sample to be predicted using the target tree, and the number of prediction trees in the store sales revenue model, the final predicted sales revenue of the sample to be predicted is determined, thus completing the prediction of store sales revenue and realizing store sales revenue data processing.

[0007] The beneficial effects are as follows: By evaluating the accuracy of each decision tree, the problem of diluting the contribution of high-accuracy trees caused by simple equal-weight averaging is avoided, thereby improving the overall accuracy of the prediction model; by combining the performance of the feature values ​​corresponding to the split points of each decision tree and the similarity between the training samples and the samples to be predicted, the weight factors are dynamically allocated, so that the decision tree more suitable for the current sample features has a greater weight in the prediction, improving the reliability of the prediction; by considering the coverage of the reference samples and the training quality of the target tree, the model has a stronger adaptability to different scenarios (such as major promotional activities and seasonal fluctuations), reducing errors; by reasonably allocating the weights of the decision trees, the impact of inefficient or irrelevant decision trees in the model on the final result is reduced, and training data and model resources are effectively utilized; more accurate and reliable store sales prediction helps enterprises to formulate more scientific inventory management, marketing strategies and resource allocation, and maximize benefits.

[0008] Furthermore, the actual sales revenue refers to the actual sales revenue of the corresponding store for each reference sample.

[0009] Furthermore, the repeated reference samples are the reference samples that appear repeatedly when selecting samples using the bootstrap sampling method during the construction of the prediction tree.

[0010] Furthermore, the aforementioned related operational features include store foot traffic data, store area data, and store promotion budget data.

[0011] Furthermore, the accuracy is obtained by: calculating the accuracy of the target tree based on the ratio of the number of repeated reference samples in the target tree to the number of reference samples selected in the target tree, and the sum of the differences between the predicted sales and the actual sales obtained by using the target tree for each reference sample in all unselected reference samples of the target tree. The accuracy is the negative correlation mapping result of the product of the ratio and the sum of the differences.

[0012] The beneficial effects are as follows: By using an exponential function to combine the proportion of repeated reference samples and the sum of errors of unselected samples, the prediction accuracy of the decision tree can be quantified, which can more accurately reflect the actual performance of each tree; by introducing the ratio of the number of repeated reference samples to the total number of selected samples, the quality of training sample coverage is reflected, avoiding the risk of overfitting due to sample duplication and improving the judgment of the model's generalization ability; by weighting the prediction error of unselected samples, the model's performance on samples outside of training is effectively captured, which helps to prevent the deterioration of predictions of unknown data by a single prediction tree.

[0013] Furthermore, the number of selected reference samples is a preset value, which is less than the number of reference samples.

[0014] Furthermore, the weight factor is obtained by: based on the accuracy, the sum of the normalized results of the values ​​of the sample to be predicted in the relevant operational features corresponding to all split points of the target tree and the normalized results of the values ​​of all split points of the target tree in the corresponding relevant operational features, and the number of reference samples that are from the same path as the sample to be predicted during the training process of the target tree, calculating the weight factor of the sample to be predicted when using the target tree for prediction, wherein the weight factor is the normalized result of the product of the accuracy, the sum of the differences, and the number.

[0015] The beneficial effects are as follows: the weighting factor combines the accuracy of the prediction tree, the matching degree between the feature values ​​of the sample to be predicted and the segmentation point, and the number of reference samples from the same source as the path of the sample to be predicted during training, thus achieving a comprehensive balance of multiple features and enhancing the accuracy of weight assignment; by exponentially decaying the normalized difference between the feature values ​​of the sample to be predicted and the corresponding segmentation point feature values, the similarity between the sample and the training path of the decision tree is reflected, ensuring that similar samples are assigned higher weights, thereby improving the relevance and accuracy of the prediction results.

[0016] Furthermore, the reference sample that is of the same origin as the sample to be predicted is a reference sample that has the same prediction path as the sample to be predicted during the training process of the target tree.

[0017] The beneficial effects are as follows: defining reference samples that are from the same path as the sample to be predicted as samples that have traversed the same prediction path during the training of the target tree can accurately identify training data that are highly consistent with the sample to be predicted in the feature space and decision path, ensuring that the selected samples have strong representativeness and relevance; using the number of reference samples from the same path as the path as an important basis for the calculation of weight factors makes the decision tree that is most experienced and targeted to the current prediction scenario in the training process have a greater influence, avoiding the waste of information caused by blindly and equally distributing weights.

[0018] Furthermore, the final predicted sales amount is obtained by taking the weight of the sample to be predicted when using each prediction tree in the store sales model, and the sum of the weights of the sample to be predicted when using all prediction trees in the store sales model, as the weight. The predicted sales amount of the sample to be predicted is then weighted and averaged using the target tree to obtain the final predicted sales amount of the sample to be predicted.

[0019] The beneficial effects are as follows: The final sales forecast uses a weighted average method, normalizing and weighting the output of each forecast tree according to its weight factor. This fully leverages the advantages of high-weight forecast trees and avoids the negative impact of low-quality forecast trees on the forecast results that may be caused by traditional equal-weighted averaging. By summing and normalizing the weight factors, the sum of all weights is ensured to be 1, making the contribution ratio of each tree relatively stable and reasonable, and preventing deviations in forecast results caused by excessively large or small individual weights. Accurate sales forecasting can help enterprises to conduct inventory management, marketing strategies, and resource allocation more scientifically, thereby improving overall operational efficiency and market competitiveness.

[0020] Secondly, the present invention provides a store sales data processing system, which adopts the following technical solution:

[0021] A store sales data processing system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned store sales data processing method is implemented.

[0022] By adopting the above technical solution, a computer program is generated from the above-mentioned method for processing store sales data and stored in a memory so that it can be loaded and executed by a processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0023] The present invention has the following technical effects:

[0024] (1) Breaking through the limitations of the traditional random forest with equal weights, by comprehensively considering multiple factors such as the training sample coverage, prediction error of unselected samples, and homology of sample paths of a single decision tree, the weight factor of each tree is dynamically allocated, so that the prediction results of the high-accuracy decision tree can be strengthened when facing complex scenarios such as major promotional activities and seasonal fluctuations, avoiding the interference of low-quality predictions on the final results, and significantly improving the accuracy and reliability of store sales forecasts.

[0025] (2) The weight adjustment mechanism based on the path homology between the sample to be predicted and the training sample can accurately identify the applicability of the decision tree to a specific sample. When a decision tree frequently encounters data patterns similar to the sample to be predicted during training, it is given higher weights to ensure that the model output is more in line with the actual business scenario and effectively reduce the prediction bias caused by sample differences.

[0026] (3) By analyzing data details such as the number of duplicate samples and the differences in features of unselected samples in the training samples, we can fully explore the hidden information relationships behind the data. This not only optimizes the decision tree weights, but also helps enterprises identify key operational characteristics and business rules, providing stronger support for operational decisions such as precision marketing and inventory management.

[0027] (4) The store sales forecasting model constructed can quickly adapt to changes in data distribution and expansion of new business scenarios by dynamically adjusting the weights, thereby reducing the risk of model overfitting. Compared with traditional methods, it has stronger generalization performance and can stably output more accurate forecast results in different stores and at different time periods, helping enterprises to achieve refined operation and cost reduction and efficiency improvement. Attached Figure Description

[0028] Figure 1 This is a flowchart of a method for processing store sales data according to an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] This invention discloses a method for processing store sales data, referring to... Figure 1 This includes steps S1-S5:

[0031] S1: Obtain the reference sample and the actual sales of the reference sample for each store, as well as the sample to be predicted for which store sales need to be predicted.

[0032] The implementers can set the number of reference samples according to the specific implementation situation, for example, 150.

[0033] The reference sample and the sample to be predicted contain multiple relevant operational characteristics.

[0034] Specifically, the actual sales revenue refers to the actual sales revenue of the store corresponding to each reference sample.

[0035] Specifically, the multiple relevant operational characteristics include store foot traffic data, store area data, and store promotion budget data.

[0036] S2: Train the random forest algorithm and use the trained model for prediction.

[0037] It should be noted that reference samples containing multiple relevant operational features and the actual sales revenue corresponding to each reference sample are input into the random forest algorithm for model training. Subsequently, the samples to be predicted, containing multiple relevant operational features, are used as input to the trained model.

[0038] The random forest algorithm is trained based on reference samples to obtain a store sales prediction model. This model is then used to predict the samples to be predicted, as follows:

[0039] S3: Denote any prediction tree in the trained model as the target tree, and determine the accuracy of the target tree.

[0040] It's important to note that when analyzing the numerical characteristics of each prediction tree to determine its accuracy, we can first analyze the degree of repetition of reference samples within each tree. High repetition of reference samples within a tree may lead to poor generalization and limited learning ability, resulting in lower accuracy. However, simply quantifying the repetition of reference samples within a single tree cannot fully reflect the model's performance. Therefore, we can also introduce the analysis of errors from unselected reference samples to objectively evaluate the performance of each prediction tree under different data distributions, avoiding reliance solely on training data for evaluation and thus arriving at more accurate and reliable conclusions. Therefore, the larger the prediction error of a prediction tree for its corresponding unselected reference samples, the greater the credibility of the high degree of sample repetition in that prediction tree, indicating poorer learning and generalization abilities, and lower accuracy.

[0041] The accuracy of the target tree is determined by the number of repeated reference samples in the target tree's selected reference samples, the number of reference samples selected by the target tree, the number of reference samples not selected by the target tree, and the difference between the predicted sales and actual sales obtained by each reference sample in the unselected reference samples of the target tree.

[0042] Specifically, the repeated reference samples are the reference samples that appear repeatedly when selecting samples using the bootstrap sampling method during the construction of the prediction tree.

[0043] Specifically, the accuracy level satisfies:

[0044] ;

[0045] In the formula, For the first The accuracy of the prediction tree, For the first The number of reference samples that are repeated among the selected reference samples in the prediction tree. The number of reference samples selected for the prediction tree. For the first The number of unselected reference samples in each prediction tree. For the first The first prediction tree in the unselected reference sample The reference sample in using the first The difference between the predicted sales obtained from the prediction tree and the actual sales obtained in advance. For hyperparameters, It is a natural exponential function.

[0046] Implementers can set hyperparameters according to the specific implementation situation, for example, 0.001. The existence of hyperparameters is to prevent... Sometimes, situations arise that render the final result meaningless.

[0047] Specifically, the number of selected reference samples is a preset value (for example, 100), which is less than the number of reference samples.

[0048] in, The larger the value, the more likely it is to be the first. The greater the repetition of reference samples within a prediction tree, the worse its generalization ability and the more limited its learning ability will be, and the lower its accuracy will be. The larger it is, the more likely it is to be the first The larger the prediction error of the unselected reference samples in the prediction tree, the more it indicates that the prediction error of the first tree is greater. The greater the degree of sample repetition in each prediction tree, the higher its reliability. The worse a prediction tree's learning and generalization abilities are, the lower its accuracy will be.

[0049] S4: Determine the weighting factors for the sample to be predicted when using the target tree for prediction.

[0050] It should be noted that by analyzing the differences between the sample to be predicted and each prediction tree regarding store traffic data, store area data, and store promotion budget data, the applicability of each prediction tree to the sample to be predicted is obtained. Combined with the accuracy of each prediction tree itself, the weighting factor corresponding to the prediction result of each prediction tree when predicting the sales data of the sample to be predicted is calculated. When analyzing the applicability of each prediction tree to the sample to be predicted, it can be indirectly judged by analyzing the degree of matching and path consistency between the features of the sample to be predicted and the features of the split points of the prediction tree. If the feature values ​​of the sample to be predicted are close to the feature values ​​of the tree's split points, it indicates that the decision path of the prediction tree is more reliable in predicting the sample to be predicted, and its applicability is greater. At the same time, if the prediction path of the sample to be predicted in a tree is traversed by more reference samples during the training of that prediction tree, it indicates that this prediction tree has learned the sample to be predicted more fully, the prediction result is more stable, and its applicability is greater.

[0051] The weighting factor of the sample to be predicted when using the target tree is determined based on the accuracy, the value of the relevant operational features corresponding to all split points of the target tree, the value of the relevant operational features corresponding to all split points of the target tree, and the number of reference samples that are from the same path as the sample to be predicted during the training process of the target tree.

[0052] Specifically, the weighting factor satisfies:

[0053] ;

[0054] In the formula, For the sample to be predicted, in using the first Weighting factors in prediction of each prediction tree For the first The accuracy of the prediction tree, For the first The number of split points in the prediction tree. For the sample to be predicted in the th The first prediction tree The values ​​of the relevant operational characteristics corresponding to each segmentation point For the first The first prediction tree The value of each segmentation point in the corresponding relevant operational characteristics. In the first The number of reference samples that share the same path as the sample to be predicted during the training of the prediction tree. It is a natural exponential function. For the standard normalized function, It is the absolute value symbol.

[0055] Specifically, the reference sample that is from the same path as the sample to be predicted is a reference sample that has the same prediction path as the sample to be predicted during the training process of the target tree.

[0056] in, The larger the value, the better the prediction tree's numerical characteristics are analyzed. The greater the accuracy of a prediction tree itself, the greater the weight factor corresponding to its prediction result for the sample to be predicted will be. Indicates the first The degree to which a prediction tree is suitable for the sample to be predicted; the larger this value, the better the prediction tree is. The larger the weight factor of the prediction value of the sample to be predicted, the greater the number of prediction trees; This indicates that the feature value of the sample to be predicted is related to the first... The degree of matching of feature values ​​of all split points in the prediction tree; the smaller this value, the better the match of the sample to be predicted with respect to the first split point. The value of the relevant operational feature corresponding to each split point of the prediction tree and the first... The smaller the difference between the feature values ​​of each split point in the prediction tree, the better the prediction tree. The more reliable the decision path of a prediction tree is for predicting the sample, the greater its applicability. The larger it is; This indicates that the sample to be predicted is in the th... The degree of path consistency between the predicted sample and all reference samples in the prediction tree; the larger this value, the better the predicted sample is in the [number]th [tree]. The more reference samples traversed a prediction path in a prediction tree during its training, the better the prediction path in that tree is. The more thoroughly a prediction tree learns from the prediction samples, the more stable its prediction results and the greater its applicability. The larger it is.

[0057] S5: Determine the final predicted sales amount for the sample to be predicted in order to process store sales data.

[0058] It should be noted that after the above analysis, the weighting factors of each prediction tree were obtained when predicting the sales data of the sample to be predicted. In this step, the final predicted sales of the sample to be predicted will be obtained by weighting the weighting factors and combining the prediction results of each tree when predicting the sales data of the sample to be predicted.

[0059] Based on the weighting factors, the predicted sales revenue obtained from the target tree for the sample to be predicted, and the number of prediction trees in the store sales revenue model, the final predicted sales revenue of the sample to be predicted is determined, thus completing the prediction of store sales revenue and realizing store sales revenue data processing.

[0060] Specifically, the final projected sales revenue satisfies:

[0061] ;

[0062] In the formula, The final projected sales revenue for the sample to be predicted. The number of prediction trees in the store sales model. For the sample to be predicted, in using the first Weighting factors in prediction of each prediction tree For the sample to be predicted, use the first The predicted sales revenue obtained from the prediction trees.

[0063] When predicting the sales data of the sample to be predicted, the larger the weight factor corresponding to the predicted sales of each prediction tree, the greater the accuracy of the prediction value itself, the greater its applicability to the sample to be predicted, and the greater its contribution to the final predicted sales of the sample to be predicted.

[0064] The above steps complete the optimization of the random forest algorithm. Subsequently, if it is necessary to predict the sales data of a new sample to be predicted, input the store traffic data, store area data, and store promotion budget data of the new sample to be predicted, and then follow the above steps to make the prediction.

[0065] This invention also discloses a store sales data processing system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a store sales data processing method according to the present invention.

[0066] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0067] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for processing store sales data, characterized in that, include: Obtain reference samples and actual sales of the reference samples for each store, as well as the samples to be predicted for which store sales need to be predicted. The reference samples and the samples to be predicted contain multiple relevant operational characteristics. The random forest algorithm is trained based on reference samples to obtain a store sales prediction model. This model is then used to predict the samples to be predicted, including: Any prediction tree in the store sales forecasting model is designated as the target tree. The accuracy of the target tree is determined based on the number of duplicate reference samples in the target tree's selected reference samples, the total number of reference samples selected by the target tree, the number of unselected reference samples in the target tree, and the difference between the predicted sales and actual sales for each unselected reference sample in the target tree. This includes calculating the accuracy of the target tree based on the ratio of the number of duplicate reference samples in the target tree's selected reference samples to the total number of reference samples selected by the target tree, and the sum of the differences between the predicted sales and actual sales for each unselected reference sample in the target tree. The accuracy is the negative correlation mapping result of the product of the ratio and the sum of the differences; based on the accuracy, the values ​​of the relevant operational features corresponding to all split points of the target tree for the sample to be predicted, the values ​​of the relevant operational features corresponding to all split points of the target tree, and the number of reference samples with the same path as the sample to be predicted during the training process of the target tree, the weight factor of the sample to be predicted when using the target tree for prediction is determined; based on the weight factor, the predicted sales amount obtained by the sample to be predicted using the target tree, and the number of prediction trees in the store sales amount model, the final predicted sales amount of the sample to be predicted is determined, and the prediction of store sales amount is completed to realize the store sales amount data processing.

2. The method for processing store sales data according to claim 1, characterized in that, The actual sales figures refer to the actual sales figures of the stores corresponding to each reference sample.

3. The method for processing store sales data according to claim 1, characterized in that, The relevant operational features include store foot traffic data, store area data, and store promotion budget data.

4. The method for processing store sales data according to claim 1, characterized in that, The repeated reference samples are those that appear repeatedly when selecting samples using the bootstrap sampling method during the construction of the prediction tree.

5. The method for processing store sales data according to claim 1, characterized in that, The number of selected reference samples is a preset value, which is less than the number of reference samples.

6. The method for processing store sales data according to claim 1, characterized in that, The weighting factor is obtained as follows: Based on the accuracy, the sum of the normalized results of the values ​​of the target sample in the relevant operational features corresponding to all split points of the target tree and the normalized results of the values ​​of the target sample in the corresponding relevant operational features, and the number of reference samples that are from the same path as the target sample during the training process of the target tree, the weight factor of the target sample when using the target tree for prediction is calculated. The weight factor is the normalized result of the product of the accuracy, the sum of the differences, and the number.

7. A method for processing store sales data according to claim 1 or 6, characterized in that, The reference sample that is from the same path as the sample to be predicted is a reference sample that has the same prediction path as the sample to be predicted during the training process of the target tree.

8. The method for processing store sales data according to claim 1, characterized in that, The method for obtaining the final projected sales revenue is as follows: The weight is calculated by taking the weight of the predicted sales amount obtained by the target tree for the sample to be predicted using the target tree. The weight is the ratio of the weight factor of each prediction tree in the model to the sum of the weight factors of all prediction trees in the model. The final predicted sales amount of the sample to be predicted is obtained by taking the weighted average of the predicted sales amount obtained by the target tree for the sample to be predicted.

9. A store sales data processing system, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a store sales data processing method according to any one of claims 1-8.

Images