Product sales prediction method based on DBSCAN and user group characteristics analysis

Abstract

The present invention relates to a product sales forecasting method based on DBSCAN and user group characteristic analysis, including: collecting order sample data; preprocessing the order sample data; clustering analysis of the order sample data using a density-based clustering algorithm; dividing user groups and performing user group characteristic analysis; using a differential autoregressive moving average model ARIMA to predict the purchasing possibility of the user group in the future time period; calculating the similarity of the user group according to the user group characteristic analysis result, and recommending products to the user group with similarity; and predicting the product sales in the future time period. The present invention predicts the purchasing possibility of the user group in the future time period through the ARIMA model, and calculates the similarity of the user group. According to the calculation result of the ARIMA model and the similarity of the user group, the predicted value of the product sales in the future time period is calculated, thereby improving the scientificity and accuracy of the product sales forecast, and facilitating the e-commerce platform and the product production party to prepare goods in advance.

Description

Technical Field

[0001] The present invention belongs to the field of computer information processing, and in particular relates to a product sales prediction method based on DBSCAN and user group characteristic analysis. Background Art

[0002] With the booming development of e-commerce, the number of goods on major e-commerce platforms has increased dramatically, bringing huge challenges to product sales forecasting. Accurately predicting product sales is of great significance to e-commerce operations. It can guide e-commerce platforms to prepare goods in time, improve resource utilization efficiency, and avoid inventory backlogs or out-of-stocks. However, existing product sales forecasting methods have the following problems:

[0003] 1) Existing methods mostly rely on the historical sales data of the product itself for prediction, and do not fully consider the user behavior data. Different users and user groups have different interests and purchasing power in products, which will directly affect the sales of products. Existing methods lack the mining and modeling of user characteristics, resulting in large deviations in prediction results.

[0004] 2) Existing methods predict product sales as a whole without distinguishing different user groups. Mixing all users together for modeling fails to reflect the behavioral characteristics of different user groups. This also reduces the pertinence and accuracy of the prediction.

[0005] 3) Existing methods lack consideration of the mutual influence between users. User groups with similar interests or characteristics will influence each other, thus affecting the sales forecast of the product. Existing product sales forecasting methods fail to model and apply this mutual influence relationship between user groups.

[0006] 4) In response to the massive and complex order data of e-commerce, existing methods directly use raw data for modeling without preprocessing the raw order data, which is easily affected by abnormal data or noise data.

[0007] In view of the above problems, the present invention studies a product sales prediction method based on a density-based clustering algorithm (Density-Based Spatial Clustering of Applications with Noise, DBSCAN) and user group behavior feature modeling to solve the deficiencies in the prior art and improve the accuracy of the prediction. Summary of the invention

[0008] The technical problem of the present invention is that the existing sales forecasting methods for e-commerce products lack consideration of the mutual influence between users / user groups, and thus cannot accurately predict product sales in future time periods.

[0009] The purpose of the present invention is to provide a product sales forecasting method based on DBSCAN and user group characteristic analysis in order to solve the above problems, to perform cluster analysis on order sample data, to divide user groups based on the cluster analysis results, to perform user group characteristic analysis, and to further calculate the similarity of user groups; to use a differential autoregressive integrated moving average model (ARIMA) to predict the purchasing possibility of user groups in future time periods; to calculate the predicted value of product sales in future time periods based on the calculation results of the ARIMA model and the similarity of user groups, to improve the scientificity and accuracy of product sales forecasting, and to facilitate e-commerce platforms and product manufacturers to prepare stocks in advance.

[0010] The technical solution of the present invention is a product sales prediction method based on DBSCAN and user group characteristic analysis, comprising the following steps:

[0011] Step 1: Collect historical order data information of users in the e-commerce mall to form order sample data;

[0012] Step 2: Preprocessing the order sample data, including data cleaning and formatting;

[0013] Step 3: Use density-based clustering algorithm to perform cluster analysis on order sample data;

[0014] Step 4: Divide the user groups according to the cluster analysis results of step 3, and perform user group characteristics analysis;

[0015] Step 5: Use the difference autoregressive moving average model ARIMA to predict the purchase possibility of the user group in the future time period;

[0016] Step 6: Based on the user group characteristic analysis results of step 4, calculate the similarity of the user groups, and combine the results of step 5 to make product recommendations for user groups with similarity;

[0017] Step 7: Based on the calculation results of steps 5 and 6, predict the product sales volume in the future time period.

[0018] Preferably, step 2 specifically includes:

[0019] 1) Identify the orders in the order sample dataset that are missing key fields such as UserID, ProductID, etc., and remove these order records from the order sample dataset;

[0020] 2) Check whether there are completely or partially identical order records in the order sample data set, and remove duplicate order records;

[0021] 3) Check and process invalid or abnormal orders where the purchase quantity in the order is less than or equal to zero or the product number ProductID does not exist.

[0022] Furthermore, step 2 also includes identifying, judging and removing the following abnormal order records:

[0023] a) Order records with abnormally high or low purchase quantities, where the purchase quantity of a certain product in the order is much higher than the average purchase quantity of the user or similar users;

[0024] b) A large number of identical order records appearing in a short period of time;

[0025] c) Order records with inconsistent geographic locations between the first and second purchases within a short period of time;

[0026] d) Order records with unknown or illegal product ID ProductID.

[0027] Furthermore, in step 3, the density-based clustering algorithm specifically includes:

[0028] 1) Represent sample data as points in space;

[0029] 2) Randomly select a sample point that has not been visited;

[0030] 3) Calculate the neighborhood of the point;

[0031] 4) Count the number of sample points in the neighborhood of the point, and compare the number of sample points with the minimum number of included points MinPts. If the number of sample points in the neighborhood is greater than or equal to the minimum number of included points MinPts, create a new cluster, otherwise mark the point as noise;

[0032] 5) Repeat steps 2) to 4) until all sample points are visited.

[0033] Preferably, in step 4, the user group characteristic analysis includes analyzing the average purchase frequency of the user group for the product, the types of products purchased, and the purchase time distribution of the product.

[0034] Furthermore, step 5 specifically includes the following sub-steps:

[0035] Step 5.1: Based on the sales records of the mall products, construct an ARIMA model for the user groups of each product to predict the purchase possibility of the user in the future time period;

[0036] Step 5.2: For the ARIMA model in step 5.1, extract relevant factors, test the correlation of relevant factors, and extract relevant factors with high correlation as independent variables for multiple linear regression;

[0037] Step 5.3: Extract autoregressive variables, build autoregressive series, differentiate the series, and further determine the order of the ARIMA model based on the autocorrelation and partial correlation parameters;

[0038] Step 5.4: Fit, verify and adjust the ARIMA model obtained in step 5.3, test the significance of each parameter, and obtain a complete ARIMA model of the product;

[0039] Step 5.5: Use the ARIMA model of the product in step 5.4 to predict the likelihood of the user group purchasing the product in the future based on the product's historical sales data;

[0040] Step 5.6: Repeat steps 5.1-5.5 to build an ARIMA model for each product in the mall, and use the ARIMA model to calculate the purchase possibility score of the user group of the product in the future period.

[0041] Furthermore, step 6 specifically includes the following sub-steps:

[0042] Step 6.1: Arrange the historical order data of users in each user group;

[0043] Step 6.2: Build a user-product matrix;

[0044] Step 6.3: Calculate the user similarity between user groups based on the user-product matrix and the user group characteristic analysis results;

[0045] Step 6.4: Based on the user similarities between user groups calculated in step 6.3, if a user group purchases a product, product recommendations are pushed to users in other user groups with similarities.

[0046] Preferably, in step 7, the predicted sales volume of the product for the user group in the future period is:

[0047] Predicted_Sales=ARIMA_Forecast×P

[0048] Where ARIMA_Forecast represents the purchase possibility score of the user group in the future period calculated by the ARIMA model, P represents the average number of users in the user group who purchased the product in the same period in history, and Predicted_Sales represents the predicted sales volume of the user group in the future period.

[0049] Considering the impact of product recommendations on product sales, the predicted sales are adjusted according to the similarity of user groups to obtain the adjusted predicted sales Adjusted_Sales.

[0050] Adjusted_Sales=Predicted_Sales×(1+Similarity_Score×w2 / w1)

[0051] Where Similarity_Score represents the similarity between the second user group outside the current user group and the current user group, w1 is the weight of the current user group, and w2 is the weight of the second user group outside the current user group;

[0052] Add up the adjusted sales forecasts for each user group to get the total sales forecast

[0053] Total_Predicted_Sales=∑Adjusted_Sales

[0054] Where Total_Predicted_Sales represents the total predicted sales of the product.

[0055] Compared with the prior art, the beneficial effects of the present invention include:

[0056] 1) The present invention predicts the purchasing possibility of a user group in a future time period through the differential autoregressive moving average model ARIMA, and calculates the similarity of the user group. According to the calculation results of the ARIMA model and the similarity of the user group, the predicted value of the product sales in the future time period is calculated, which solves the problem that the existing product sales forecasting method lacks consideration of the mutual influence between users and leads to inaccurate forecasting. The present invention improves the scientificity and accuracy of product sales forecasting, and is convenient for e-commerce platforms and product manufacturers to prepare stocks in advance.

[0057] 2) The present invention uses a density-based clustering algorithm to cluster a large amount of historical order sample data, i.e., a method of big data analysis, to divide user groups and classify users with similar characteristics into the same category, so as to facilitate accurate product sales forecasting based on user groups; the density-based clustering algorithm has a good denoising effect and is used for cluster analysis of user groups, which can avoid the influence of abnormal order data on cluster analysis results and realize accurate distinction of user groups;

[0058] 3) Before using order sample data to predict the purchasing possibility of user groups in future time periods, the method of the present invention cleans and formats the order sample data, and identifies, judges and removes order records with abnormal conditions, effectively avoiding the impact of erroneous and abnormal order data on the prediction results, and further improving the accuracy of product sales prediction results.

[0059] 4) The present invention is based on the user-product matrix and combines user characteristics such as the average purchase frequency of products by the user group, the types of products purchased, and the purchase time distribution of products to calculate the similarity of the user groups, making the measurement of the similarity of the user groups more scientific and reasonable.

[0060] 5) The present invention achieves effective distinction between primary and secondary user groups of a product by setting different weights for different user groups, which is conducive to accurate prediction of product sales. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0062] Figure 1 It is a flowchart of the product sales forecasting method of the present invention.

[0063] Figure 2 Flow chart of a product sales forecasting method according to an embodiment of the present invention.

[0064] Figure 3 It is a timing diagram of the DBSCAN clustering algorithm in an embodiment of the present invention.

[0065] Figure 4 It is a time series diagram of the ARIMA model used for prediction in an embodiment of the present invention.

[0066] Figure 5 This is a diagram of an order statistics interface of an operation analysis system applied to e-commerce according to an embodiment of the present invention. DETAILED DESCRIPTION

[0067] like Figure 5 As shown, in the embodiment, an operation analysis system applied to e-commerce is constructed. By deeply analyzing the user's purchase orders, the user characteristics are better understood. According to the user similarity, product purchase suggestions are pushed to similar users, so as to give full play to the mutual influence between similar users.

[0068] like Figure 1 and Figure 2 As shown, the product sales prediction method based on DBSCAN and user group characteristic analysis includes the following steps:

[0069] Step 1: Collect historical order data information of users in the e-commerce mall to form order sample data;

[0070] In the embodiment, the order data includes the user number UserID, the order number OrderID, the product number ProductID, the purchase quantity Quantity, the purchase time OrderTime and the geographical location information UserLocation where the user places the order.

[0071] Step 2: Preprocessing the order sample data, including data cleaning and formatting;

[0072] Data cleaning of the order sample data set includes:

[0073] 1) Identify the orders in the order sample dataset that are missing key fields such as UserID, ProductID, etc., and remove these order records from the order sample dataset;

[0074] 2) Check whether there are completely or partially identical order records in the order sample data set, and remove duplicate order records;

[0075] 3) Check and process invalid or abnormal orders where the purchase quantity is less than or equal to zero or the product ID ProductID does not exist.

[0076] In the embodiment, step 2 also includes identifying, judging and removing the order records of the following abnormal situations:

[0077] a) Order records with abnormally high or low purchase quantities, where the purchase quantity of a certain product in the order is much higher than the average purchase quantity of the user or similar users;

[0078] b) A large number of identical order records appearing in a short period of time;

[0079] c) Order records with inconsistent geographic locations between the first and second purchases within a short period of time;

[0080] d) Order records with unknown or illegal product ID ProductID.

[0081] In the embodiment, manual review is performed on the following situations:

[0082] ① Abnormally high or low purchase quantity: If the purchase quantity of a product in an order is much higher or lower than the average purchase quantity of this user or similar users, this is a situation that requires further manual review. Such anomalies may be caused by data entry errors or other unknown factors.

[0083] ② A large number of identical orders in a short period of time: If a large number of almost identical orders are generated under the same user ID within a very short period of time, such as a few minutes or seconds, this may be caused by system errors or malicious behavior.

[0084] ③Inconsistent geographic location information: If the same user ID reports different geographic location information within a short period of time, such as switching from New York to Tokyo within a few minutes, this is also worthy of manual review.

[0085] ④Unknown or illegal ProductID: If a ProductID that does not exist in the database appears in the order, or the ProductID obviously does not match the user's purchase history, for example, a children's toy ProductID appears in an industrial equipment purchase order.

[0086] Step 3: Use density-based clustering algorithm to perform cluster analysis on order sample data, such as Figure 3 shown.

[0087] In the embodiment, the specific process of the density-based clustering algorithm includes:

[0088] 1) Represent sample data as points in space;

[0089] Convert the order sample data into points that can be represented in multidimensional space. Each order sample consists of multiple attributes such as purchase amount, purchase time, and type of goods purchased, which can be regarded as coordinates in multidimensional space.

[0090] In the embodiment, the order sample data is obtained from the database and stored in the form of a NumPy array. Each row represents a user order sample, and each column represents a feature such as purchase amount, purchase time, and type of purchased goods. This representation facilitates the use of spatial distance metrics to evaluate the similarity between sample points.

[0091] 2) Randomly select a sample point that has not been visited;

[0092] Randomly select a sample point from the order sample data set that has not been visited or considered. To achieve this goal, first maintain a set of indices of "visited" sample points. This set is empty at the beginning of the algorithm and is updated at each iteration; then calculate the set of indices of sample points that have not been visited, which is done by subtracting the set of indices of "visited" sample points from the set of indices of the total sample points; then, randomly select an index from this set of "unvisited" indices.

[0093] Randomly selecting unvisited sample points means that the algorithm does not rely on the initial ordering of the dataset, which increases the robustness of the algorithm. By randomly selecting sample points, it ensures that the algorithm can fully and evenly explore the space of the dataset, thereby more accurately identifying user groups with similar characteristics. This also provides a basis for subsequent steps, especially in calculating the neighborhood of the sample points and evaluating their density.

[0094] 3) Calculate the neighborhood of the point;

[0095] In a multidimensional data space, each data point has an area around it called a "neighborhood".

[0096] The radius∈ is a predefined parameter that determines when two points are considered "close". The choice of its value usually depends on the specific characteristics of the dataset and the application scenario. Different values ​​of radius∈ may lead to very different clustering results.

[0097] After determining the distance metric and radius ∈, the algorithm generates a set of neighborhood points. This set is actually an index list pointing to all points within the distance ∈ from the currently selected point.

[0098] 4) Count the number of sample points in the neighborhood of the point, and compare the number of sample points with the minimum number of included points MinPts. If the number of sample points in the neighborhood is greater than or equal to the minimum number of included points MinPts, create a new cluster, otherwise mark the point as noise;

[0099] The minimum number of included points MinPts is a predefined parameter that is determined by the nature of the dataset and specific business requirements. This parameter will be used to determine whether the neighborhood of a point is "dense" enough to form a cluster. Count the number of sample points in the neighborhood. After obtaining the neighborhood of a point, the next step is to count how many sample points there are in this neighborhood.

[0100] If the number of sample points in the neighborhood is greater than or equal to MinPts, then this point will be used as a core point of a new cluster. If the number of sample points in the neighborhood is less than MinPts, then this point will be marked as noise.

[0101] 5) Repeat steps 2) to 4) until all sample points are visited.

[0102] In the embodiment, the code for traversing all sample points is as follows:

[0103]

[0104] Step 4: Divide the user groups according to the cluster analysis results of step 3, and perform user group characteristics analysis;

[0105] In the embodiment, the average purchase frequency of products, the types of products purchased, and the purchase time distribution of products by the user group are analyzed.

[0106] Step 5: Use the difference autoregressive moving average model ARIMA to predict the purchase possibility of the user group in the future time period, such as Figure 4 shown.

[0107] Step 5.1: Based on the product sales records in the order sample data, construct an ARIMA model for each product user group to predict the user's purchase possibility in the future time period;

[0108] Step 5.2: For the ARIMA model in step 5.1, extract relevant factors, test the correlation of relevant factors, and extract relevant factors with high correlation as independent variables for multiple linear regression;

[0109] Step 5.3: Extract autoregressive variables, build autoregressive series, differentiate the series, and further determine the order of the ARIMA model based on the autocorrelation and partial correlation parameters;

[0110] Step 5.4: Fit, verify and adjust the ARIMA model obtained in step 5.3, test the significance of each parameter, and obtain a complete ARIMA model of the product;

[0111] Step 5.5: Use the ARIMA model of the product in step 5.4 to predict the probability score of the user group purchasing the product in the future based on the historical sales data of the product;

[0112] Step 6: Based on the user group characteristic analysis results of step 4, calculate the similarity of the user groups, and combine the results of step 5 to make product recommendations for user groups with similarity;

[0113] Step 6.1: Sort out the order data of each user in the order sample data set;

[0114] Step 6.2: Build a user-product matrix;

[0115] The user-product matrix provides a structured data form for subsequent analysis. Each row represents a user, each column represents a product, and each value in the matrix represents the number of purchases or other attention indicators of the user for the product.

[0116] In the embodiment, the code for implementing the user-product matrix is ​​as follows:

[0117]

[0118] Step 6.3: Calculate the user similarity between user groups based on the user-product matrix and the user group characteristic analysis results;

[0119] Step 6.3.1: Use the user-product matrix to calculate user similarity based on purchase history;

[0120] Step 6.3.2: Use the user group feature vector to calculate the similarity based on user features.

[0121] Step 6.3.3: Combine the above two similarities to get the final similarity of the user group.

[0122] In the embodiment, a weighted average calculation method is adopted to calculate the comprehensive user similarity based on the user similarity index of the purchase history and the similarity index based on the user characteristics, and the similarity of the user group is further calculated based on the comprehensive user similarity.

[0123] Step 6.4: Based on the user similarities between user groups calculated in step 6.3, if a user group purchases a product, product recommendations are pushed to users in other user groups with similarities.

[0124] Step 7: Based on the calculation results of steps 5 and 6, predict the product sales volume in the future time period.

[0125] The predicted sales volume of the product for the user group in the future period is:

[0126] Predicted_Sales=ARIMA_Forecast×P

[0127] Where ARIMA_Forecast represents the purchase possibility score of the user group in the future period calculated by the ARIMA model, P represents the average number of users in the user group who purchased the product in the same period in history, and Predicted_Sales represents the predicted sales volume of the user group in the future period.

[0128] Considering the impact of product recommendations on product sales, the predicted sales are adjusted according to the similarity of user groups to obtain the adjusted predicted sales Adjusted_Sales.

[0129] Adjusted_Sales=Predicted_Sales×(1+Similarity_Score×w2 / w1)

[0130] Where Similarity_Score represents the similarity between the second user group outside the current user group and the current user group, w1 is the weight of the current user group, and w2 is the weight of the second user group outside the current user group;

[0131] Add up the adjusted sales forecasts for each user group to get the total sales forecast

[0132] Total_Predicted_Sales=∑Adjusted_Sales

[0133] Where Total_Predicted_Sales represents the total predicted sales of the product.

[0134] In the embodiment, the order statistics interface of the operation analysis system applied to e-commerce is as follows: Figure 5As shown in the figure, the daily and monthly forecasts of valid orders are presented. Based on the results of the user group characteristics analysis, the number of valid orders per day and the daily estimated invoicing amount of the applied orders are predicted for the next 7 days. This interface intuitively displays the changes in the number of orders and the forecast trend through bar charts and line charts, allowing operators to quickly understand and analyze product sales, which helps improve operational efficiency and assists in decision-making.

[0135] In the embodiment, the operation analysis system applied to e-commerce predicts the expected order volume for each province in the next 7 days based on the results of the analysis of the order characteristics of the user group, which can help the mall better understand the order situation in each region, so as to make more accurate operation decisions. The operation analysis system applied to e-commerce displays the predicted results in the form of a map of China. The province with a larger order volume has a darker color on the map. Such a visual display can help operators more intuitively understand the predicted order volume in each region, so as to make more accurate decisions.

Claims

1. A product sales prediction method based on DBSCAN and user group characteristic analysis, characterized in that: The following steps are involved: Step 1: Collect historical order data information of users in the e-commerce mall to form order sample data; Step 2: Preprocessing the order sample data, including data cleaning and formatting; Step 3: Use density-based clustering algorithm to perform cluster analysis on order sample data; Step 4: Divide the user groups according to the cluster analysis results of step 3, and perform user group characteristics analysis; Step 5: Use the difference autoregressive moving average model ARIMA to predict the purchase possibility of the user group in the future time period; Step 5.1: Based on the sales records of the mall products, construct an ARIMA model for the user groups of each product to predict the purchase possibility of the user in the future time period; Step 5.2: For the ARIMA model in step 5.1, extract relevant factors, test the correlation of relevant factors, and extract relevant factors with high correlation as independent variables for multiple linear regression; Step 5.3: Extract autoregressive variables, build autoregressive series, differentiate the series, and further determine the order of the ARIMA model based on the autocorrelation and partial correlation parameters; Step 5.4: Fit, verify and adjust the ARIMA model obtained in step 5.3, test the significance of each parameter, and obtain a complete ARIMA model of the product; Step 5.5: Use the ARIMA model of the product in step 5.4 to predict the likelihood of the user group purchasing the product in the future based on the product's historical sales data; Step 5.6: Repeat steps 5.1-5.5 to build an ARIMA model for each product in the mall, and use the ARIMA model to calculate the purchase possibility score of the user group of the product in the future period; Step 6: Based on the user group characteristic analysis results of step 4, calculate the similarity of the user groups, and combine the results of step 5 to make product recommendations for user groups with similarity; Step 6.1: Arrange the historical order data of users in each user group; Step 6.2: Build a user-product matrix; Step 6.3: Calculate the user similarity between user groups based on the user-product matrix and the user group characteristic analysis results; Step 6.4: Based on the user similarities between the user groups calculated in step 6.3, if a user group purchases a product, product recommendations are pushed to users of other user groups with similarities; Step 7: Based on the calculation results of steps 5 and 6, predict the product sales volume in the future time period; The predicted sales volume of the product for the user group in the future period is: Predicted_Sales = ARIMA_Forecast × P ; Where ARIMA_Forecast represents the purchase possibility score of the user group in the future period calculated by the ARIMA model. P It indicates the average number of users in the user group who purchased the product in the same period in history. Predicted_Sales indicates the predicted sales volume of the product purchased by the user group in the future period. Considering the impact of product recommendations on product sales, the predicted sales are adjusted according to the similarity of user groups to obtain the adjusted predicted sales Adjusted_Sales. Adjusted_Sales = Predicted_Sales ×(1 + Similarity_Score × w 2 / w 1); Where Similarity_Score represents the similarity between the second user group outside the current user group and the current user group. w 1 is the weight of the current user group, w 2 is the weight of the second user group other than the current user group; Add up the adjusted sales forecasts for each user group to get the total sales forecast Total_Predicted_Sales =∑Adjusted_Sales; Where Total_Predicted_Sales represents the total predicted sales of the product.

2. The product sales forecasting method according to claim 1, characterized in that: Step 2 specifically includes: 1) Identify the orders with missing key fields in the order sample dataset and remove these order records from the order sample dataset; 2) Check whether there are completely or partially identical order records in the order sample data set, and remove duplicate order records; 3) Check and process invalid or abnormal orders where the purchase quantity in the order is less than or equal to zero or the product number ProductID does not exist.

3. The product sales forecasting method according to claim 2, characterized in that: Step 2 also includes identifying, judging and removing the following abnormal order records: a) Order records with abnormally high or low purchase quantities, where the purchase quantity of a certain product in the order is much higher than the average purchase quantity of the user or similar users; b) A large number of identical order records appearing in a short period of time; c) Order records with inconsistent geographic locations between the first and second purchases within a short period of time; d) Order records with unknown or illegal product ID ProductID.

4. The product sales forecasting method according to claim 1, 2 or 3, characterized in that: In step 3, the density-based clustering algorithm specifically includes: 1) Represent sample data as points in space; 2) Randomly select a sample point that has not been visited; 3) Calculate the neighborhood of the point; 4) Count the number of sample points in the neighborhood of the point, and compare the number of sample points with the minimum number of included points MinPts. If the number of sample points in the neighborhood is greater than or equal to the minimum number of included points MinPts, create a new cluster, otherwise mark the point as noise; 5) Repeat steps 2) to 4) until all sample points are visited.

5. The product sales forecasting method according to claim 4, characterized in that: In step 4, the user group characteristic analysis includes analyzing the average purchase frequency of the user group for the product, the types of products purchased, and the purchase time distribution of the product.

Images