Low-income crowd identification method and device, equipment, storage medium and product

Abstract

The invention belongs to the technical field of data processing, and discloses a low-income crowd identification method and device, equipment, a storage medium and a product. The method comprises the following steps: training a plurality of initial gradient boosting models through target feature data of low-income crowds to obtain a plurality of target gradient boosting models; integrating the plurality of target gradient lifting models to obtain a target recognition model; and inputting to-be-identified data into the target identification model to obtain a low-income crowd identification result corresponding to the to-be-identified data. According to the method, the plurality of initial gradient lifting models are respectively trained, the plurality of target gradient lifting models are integrated, the weight of each target gradient lifting model is dynamically adjusted in the integration process, and then low-income crowd identification is performed on the to-be-identified data according to the target identification model. Therefore, the accuracy of low-income crowd recognition can be improved by integrating a plurality of gradient lifting models.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, equipment, storage medium and product for identifying low-income people. Background Art

[0002] Currently, low-income people can be identified through a multiple linear regression scheme, which uses multiple factors to predict an individual's income level. However, the prediction results of this model may be affected by problems such as data outliers and collinearity, resulting in inaccurate prediction results. In addition, the model has poor interpretability, and it is difficult to understand which factors have a greater impact on the prediction results, which limits the in-depth understanding of low-income people. Therefore, how to accurately and effectively identify low-income people has become a problem that needs to be solved urgently. Summary of the invention

[0003] The main purpose of this application is to provide a method, device, equipment, storage medium and product for identifying low-income people, aiming to solve the technical problem of how to accurately and effectively identify low-income people.

[0004] To achieve the above objectives, the present application provides a method for identifying low-income people, the method comprising the following steps:

[0005] Using the target feature data of low-income people, multiple initial gradient boosting models are trained separately to obtain multiple target gradient boosting models;

[0006] Integrating the multiple target gradient boosting models to obtain a target recognition model;

[0007] The data to be identified is input into the target identification model to obtain the low-income population identification result corresponding to the data to be identified.

[0008] Optionally, the step of respectively training multiple initial gradient boosting models using target feature data of low-income people to obtain multiple target gradient boosting models specifically includes:

[0009] Performing feature processing on initial feature data corresponding to low-income people to obtain processed feature data;

[0010] Resampling the processed feature data by a preset sampling algorithm to obtain target feature data;

[0011] Dividing the target feature data to obtain a training set and a test set;

[0012] A plurality of initial gradient boosting models are trained respectively by using the training set and the test set to obtain a plurality of target gradient boosting models.

[0013] Optionally, the step of performing feature processing on the initial feature data corresponding to the low-income population to obtain the processed feature data specifically includes:

[0014] The initial characteristic data corresponding to the low-income population is screened according to the classification labels to obtain the screened characteristic data;

[0015] Performing feature binning processing on the filtered feature data to obtain binned feature data;

[0016] Perform feature cross processing on the binned feature data to obtain processed feature data.

[0017] Optionally, the step of resampling the processed feature data by a preset sampling algorithm to obtain target feature data specifically includes:

[0018] Eliminating noise samples in the processed feature data to obtain a sample set;

[0019] Selecting a plurality of minority class samples from the sample set, and clustering the plurality of minority class samples to obtain an initial minority class cluster;

[0020] Determine a target minority cluster according to the Euclidean distance between each of the initial minority clusters, and determine a sampling weight corresponding to each of the target minority clusters;

[0021] Target feature data is determined according to the sampling weights.

[0022] Optionally, the step of integrating the multiple target gradient boosting models to obtain a target recognition model specifically includes:

[0023] Determining an initial set of hyperparameters corresponding to the multiple target gradient boosting models;

[0024] Optimizing the initial hyperparameter set to obtain a target hyperparameter combination;

[0025] The multiple target gradient boosting models are integrated according to the target hyperparameter combination to obtain a target recognition model.

[0026] Optionally, the step of optimizing the initial hyperparameter set to obtain a target hyperparameter combination specifically includes:

[0027] Randomly selecting initial hyperparameter combinations corresponding to the multiple target gradient boosting models from the initial hyperparameter set;

[0028] Determine the expected value of the number of fitness evaluations corresponding to each of the initial hyperparameter combinations by a multi-starting point hill climbing algorithm;

[0029] A target hyperparameter combination is selected from the initial hyperparameter combinations according to the expected value.

[0030] In addition, to achieve the above-mentioned purpose, the present application also provides a low-income people identification device, the low-income people identification device comprising:

[0031] A model training module, used to train multiple initial gradient boosting models respectively using target feature data of low-income people to obtain multiple target gradient boosting models;

[0032] A model integration module, used to integrate the multiple target gradient boosting models to obtain a target recognition model;

[0033] The identification module is used to input the data to be identified into the target identification model to obtain the low-income population identification results corresponding to the data to be identified.

[0034] In addition, to achieve the above-mentioned purpose, the present application also proposes a device for identifying low-income people, and the device for identifying low-income people includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the method for identifying low-income people as described above.

[0035] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the method for identifying low-income people as described above are implemented.

[0036] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the method for identifying low-income people as described above.

[0037] This application trains multiple initial gradient boosting models separately through target feature data of low-income people to obtain multiple target gradient boosting models; integrates the multiple target gradient boosting models to obtain a target recognition model; inputs the data to be recognized into the target recognition model to obtain the low-income people recognition result corresponding to the data to be recognized. This application first trains multiple initial gradient boosting models separately, and integrates multiple target gradient boosting models, dynamically adjusts the weights of each target gradient boosting model during the integration process, and then identifies low-income people based on the target recognition model for the data to be recognized, thereby improving the accuracy of low-income people recognition by integrating multiple gradient boosting models. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0040] Figure 1 This is a flow chart of the first embodiment of the method for identifying low-income people of the present application;

[0041] Figure 2 A schematic diagram of the fields of communication operator data of an embodiment of a method for identifying low-income people of the present application;

[0042] Figure 3 This is a flow chart of the second embodiment of the method for identifying low-income people of the present application;

[0043] Figure 4 A correlation coefficient matrix diagram of the processed feature data of an embodiment of the low-income population identification method of the present application;

[0044] Figure 5 This is a schematic diagram of feature distribution of an embodiment of a method for identifying low-income people of the present application;

[0045] Figure 6 A schematic diagram of the comparison of positive and negative sample data in an embodiment of the method for identifying low-income people of the present application;

[0046] Figure 7 A schematic diagram of the SMOTE algorithm and the RUA-SMOTE algorithm of an embodiment of a method for identifying low-income people of the present application;

[0047] Figure 8 This is a flow chart of the third embodiment of the method for identifying low-income people of the present application;

[0048] Fig. 9 This is a Voting algorithm effect diagram of an embodiment of a method for identifying low-income people in this application;

[0049] Fig.10 This is a structural block diagram of the first embodiment of the low-income identification device of the present application;

[0050] Fig.11 It is a structural diagram of an identification device for low-income people in the hardware operating environment involved in the embodiment of the present application.

[0051] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0052] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0053] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0054] The main solution of the embodiment of the present application is: to train multiple initial gradient boosting models separately through the target feature data of low-income people to obtain multiple target gradient boosting models; to integrate the multiple target gradient boosting models to obtain a target recognition model; to input the data to be recognized into the target recognition model to obtain the low-income people recognition result corresponding to the data to be recognized.

[0055] Currently, low-income people can be identified through a multiple linear regression scheme, which uses multiple factors to predict an individual's income level. However, the prediction results of this model may be affected by problems such as data outliers and collinearity, resulting in inaccurate prediction results. In addition, the model has poor interpretability, making it difficult to understand which factors have a greater impact on the prediction results, limiting the in-depth understanding of low-income people.

[0056] This application trains multiple initial gradient boosting models separately through target feature data of low-income people to obtain multiple target gradient boosting models; integrates the multiple target gradient boosting models to obtain a target recognition model; inputs the data to be recognized into the target recognition model to obtain the low-income people recognition result corresponding to the data to be recognized. This application first trains multiple initial gradient boosting models separately, and integrates multiple target gradient boosting models, dynamically adjusts the weights of each target gradient boosting model during the integration process, and then identifies low-income people based on the target recognition model for the data to be recognized, thereby improving the accuracy of low-income people recognition by integrating multiple gradient boosting models.

[0057] It should be noted that the execution subject of this application can be a computing service device with data processing, network communication and program running functions, such as a computer, or an electronic device capable of realizing the above functions, a low-income people identification device, etc. The following takes the low-income people identification device as an example to illustrate this embodiment and the following embodiments.

[0058] Based on this, the present application embodiment provides a method for identifying low-income people. Figure 1 , Figure 1This is a flow chart of the first embodiment of the method for identifying low-income people in the present application.

[0059] In this embodiment, the method for identifying low-income people includes the following steps:

[0060] Step S10: training multiple initial gradient boosting models respectively using target feature data of low-income people to obtain multiple target gradient boosting models.

[0061] It is understandable that the target feature data may include data provided by relevant units and communication operator data. The data provided by relevant units are files of poor students obtained and sorted by superior units. The files contain attributes such as name, gender, ethnicity, ID number, date of birth, registration number, political status, education level, per capita net income of the family, poverty attributes, etc. This part of the data is matched and verified in the system of relevant units. Communication operator data may include 13 indicators such as car ownership, house ownership, monthly phone bill amount, disability status, provident fund status, monthly call time, medical treatment tendency, consumer APP, consumption level, poor student tendency, employment status, social assistance tendency, and terminal equipment model. The main fields are as follows: Figure 2 As shown, Figure 2 This is a schematic diagram of the fields of communication operator data in an embodiment of the method for identifying low-income people of the present application.

[0062] It should be understood that the initial gradient boosting model may include models such as LightGBM, XGBoost, and GradientBoosting. In this embodiment, each initial gradient boosting model may be trained separately using target feature data of low-income people to obtain multiple target gradient boosting models.

[0063] Step S20: Integrate the multiple target gradient boosting models to obtain a target recognition model.

[0064] It is understandable that the present embodiment can integrate multiple target gradient boosting models, and can adjust the weight reference of each target gradient boosting model during the integration process to obtain a target recognition model.

[0065] Step S30: input the data to be identified into the target identification model to obtain the low-income population identification result corresponding to the data to be identified.

[0066] It should be understood that the data to be identified refers to data that needs to be identified as to whether the corresponding person is a low-income person. It can be data such as car ownership, house ownership, monthly phone bills, disability status, etc. By inputting the data to be identified into the target recognition model, the low-income person identification result can be obtained, that is, whether the person corresponding to the data to be identified is a low-income person.

[0067] In this embodiment, multiple initial gradient boosting models are trained separately through target feature data of low-income people to obtain multiple target gradient boosting models; the multiple target gradient boosting models are integrated to obtain a target recognition model; the data to be recognized is input into the target recognition model to obtain the low-income people recognition result corresponding to the data to be recognized. In this embodiment, multiple initial gradient boosting models are trained separately first, and multiple target gradient boosting models are integrated, and the weights of each target gradient boosting model are dynamically adjusted during the integration process, and then low-income people are recognized for the data to be recognized according to the target recognition model, so that the accuracy of low-income people recognition can be improved by integrating multiple gradient boosting models.

[0068] refer to Figure 3 , Figure 3 This is a flow chart of the second embodiment of the method for identifying low-income people in the present application.

[0069] Based on the above first embodiment, in this embodiment, step S10 includes:

[0070] Step S101: performing feature processing on initial feature data corresponding to low-income people to obtain processed feature data.

[0071] It is understandable that the discrete variables in this embodiment use 0ne-hot coding to preprocess the initial feature data. For example, there are currently 13 prefecture-level cities in Jiangsu Province that have preprocessed data to form City_1, City_2, ..., City_13, a total of 13. To further study the relationship between user characteristics and whether they are low-income, calculate the correlation coefficient and p-value based on the constructed main field contingency table, and judge whether there are differences between different behaviors:

[0072]

[0073] Among them, the p-value is a parameter used to determine the result of the hypothesis test. It can be calculated and compared using the rejection domain of the distribution according to different distributions. A small p-value indicates that the probability of the occurrence of the null hypothesis is very small. According to the principle of small probability, there is reason to reject the null hypothesis. The smaller the p-value, the more sufficient the reason for rejecting the null hypothesis. f0 is the actual frequency of the sample, f e is the expected sample frequency, is the correlation coefficient, χ 2 It is the degree of deviation between the actual observed value and the theoretical inference value.

[0074] For example, the null hypothesis H0: whether there is no difference between the labor force and the low-income group; the alternative hypothesis H1: whether there is a difference between the labor force and the low-income group. Through Python calculation, χ 2 =452.7, p = 0.00012. Therefore, under the condition of significance level a = 0.05, the null hypothesis should be rejected, and it can be known whether there are differences between the labor force and the low-income group. In addition, from the perspective of data rationality, the number of labor force and retired people is very different. Separate modeling can also effectively reduce the error caused by data imbalance, which is also a unique feature.

[0075] Furthermore, in order to effectively perform feature processing on the initial feature data, in this embodiment, step S101 includes: screening the initial feature data corresponding to the low-income population according to the classification label to obtain screened feature data; performing feature binning processing on the screened feature data to obtain binned feature data; performing feature cross-processing on the binned feature data to obtain processed feature data.

[0076] It is understandable that 0ne-hot encoding is used to represent discrete classification labels, and independent labels are represented as independent numbers, and the distances between the numbers are also equal; after univariate test screening, data unrelated to user characteristics and labels are eliminated to obtain the screened feature data.

[0077] It should be understood that in the filtered feature data, the attributes involved are all continuous variables. In some cases, a slight change in the feature value may have a significant impact on the model results. In order to improve the stability of the model and avoid the influence of outliers, the filtered feature data can be binned and divided into 6 equal-proportional intervals.

[0078] In specific implementations, the ability of machine learning algorithms to fit nonlinear patterns usually determines the accuracy that the algorithm can achieve. In order to enhance the algorithm's ability to fit feature nonlinear patterns, a feature crossover method is used to combine two or more input features to express the nonlinear laws of the features. For example, if the average consumption of a package is 80 yuan each time, the results of judging whether the labor force and the retired population are poor may be different. Therefore, two different features, "labor force-80" and "retirement-80", can be constructed, and the original feature data after binning are crossovered to form the processed feature data.

[0079] In addition, the python modules matplotlib and seaborn are used for mapping, and the correlation coefficient heat map is drawn for the processed feature data. The heat map of the correlation coefficient between features is displayed. Through the heat map, the correlation structure between the features in the data can be intuitively discovered. This helps to gain a deeper understanding of the patterns and characteristics of the data, and provides guidance for subsequent data analysis, visualization, and modeling, and the degree of correlation between the main indicators. Figure 4This is a correlation coefficient matrix diagram of the processed feature data of an embodiment of the low-income population identification method of this application, Figure 5 This is a characteristic distribution diagram of an embodiment of a method for identifying low-income people in this application. Figure 6 This is a schematic diagram of the comparison of positive and negative sample data in an embodiment of the low-income population identification method of the present application. Table 1 shows the mean values ​​of each indicator of the processed feature data.

[0080] Table 1

[0081] type Number of outgoing calls Caller Talk Time Number of callers 0 89.5 8408.199 25.96 1 38.1 3188.34 11.12 type Days of use of travel apps Number of times travel apps were used Traffic usage by travel apps 0 0.36 0.32 354198.8926 1 0.16 0.12 33370.98914 type Is the labor force ratio Roaming days SMS volume 0 0.87 2.18 86.27 1 0.48 0.34 70.53 type Days of use of consumer apps Number of times consumer apps are used Data usage by consumer apps 0 6.04 10.99 47111034.03 1 3.28 5.26 17005267.45 type Number of days of using banking apps Number of times banking apps are used Data usage of banking apps 0 0.045 0.024 21785.32 1 0.014 0.001 1049.18 type Days of use of food delivery apps Number of times food delivery apps are used Data usage of food delivery apps 0 1.29 0.42 121390.2967 1 0.53 0.15 84942.41877 type User age Car ownership ratio Average consumption level 0 41.62 0.63 2.4 1 57.43 0.13 1.7 type Working ratio is the proportion of urban households Provident Fund SMS Number 0 0.87 0.4 0.028 1 0.52 0.31 0.00584

[0082] Step S102: resampling the processed feature data using a preset sampling algorithm to obtain target feature data.

[0083] Furthermore, in order to effectively resample the processed feature data, in this embodiment, the step S102 includes: removing noise samples in the processed feature data to obtain a sample set; selecting multiple minority class samples from the sample set, and clustering the multiple minority class samples to obtain initial minority class clusters; determining the target minority class cluster according to the Euclidean distance between each of the initial minority class clusters, and determining the sampling weight corresponding to each of the target minority class clusters; determining the target feature data according to the sampling weights.

[0084] It is understandable that due to the relatively scattered distribution of low-income people and the opaque information in the scene of low-income people identification after the processed feature data, the collection of positive samples is more difficult and costly. Secondly, low-income people often lack Internet access or use digital devices less frequently, which increases the difficulty of collecting data through online platforms. There is also the imbalance of data distribution. Unbalanced data distribution will cause the model to be more likely to predict the majority class, that is, negative samples, during the training process, thereby reducing the accuracy of identifying low-income people and the generalization ability of the model. For the classification problem of unbalanced data sets, the existing oversampling algorithm mainly solves the imbalance problem between categories, but does not consider the imbalance problem within the minority class. The existing algorithm does not screen the samples for oversampling, does not remove noise, and there is a sample overlap problem in the synthesis process. This embodiment proposes an oversampling algorithm (RUA-Synthetic Minority Oversampling Technique, RUA-SMOTE) that combines hierarchical clustering, random undersampling and an improved version of SMOTE. The method performs hierarchical clustering on minority class samples, and uses random under-sampling according to the imbalance ratio of the current samples, and then uses the improved SMOTE algorithm to oversample the minority class. The preset sampling algorithm in this embodiment can be the above-mentioned RUA-SMOTE algorithm.

[0085] It should be understood that reference Figure 7 , Figure 7 Schematic diagram of the SMOTE algorithm and the RUA-SMOTE algorithm of the low-income population identification method of the present application. Figure 7 As shown in (a), we first randomly select a minority class sample c, then find the k nearest neighbors of the sample, assuming k = 5, and then randomly select a sample d from the k nearest neighbors, and randomly find a point e on the line between the minority class sample c and the selected minority class sample d, which is the synthetic sample point. Obviously, sample e appears in the space of the majority class. In this case, it will not only fail to help the training of the classifier, but will make the data more difficult to distinguish because of sample e. Therefore, some measures should be taken to ensure the rationality of the synthetic sample.

[0086] The RUA-SMOTE algorithm first uses a hierarchical clustering algorithm on the minority class to divide the minority class into several clusters and synthesize new samples in each cluster. Figure 7 As shown in (b), several clusters are generated on the minority class, and then new samples are synthesized in each cluster. Taking cluster A as an example, the algorithm process is: randomly select several minority class samples, such as randomly select two sample points, point g and point h, use linear interpolation to randomly generate a point f on the line connecting the two points, and then use the newly synthesized minority class intermediate sample and the cluster center to synthesize a new sample j, and finally add j to the minority class sample set.

[0087] In a specific implementation, the noise samples in the processed feature data can be eliminated to obtain a sample set. Specifically, the idea of ​​K nearest neighbors can be used to filter the noise samples in the processed feature data I. Set K=5, traverse the samples in I. If more than 4 / 5 of the K nearest neighbors of the sample in I are of the opposite sample category of the selected sample, the sample is determined to be a noise sample and eliminated. The remaining sample points form a sample set I'.

[0088] In this embodiment, multiple minority class samples can be selected from the sample set, and the multiple minority class samples are clustered to obtain an initial minority class cluster. Specifically, the data set is adaptively randomized and under-sampled to reduce the number of majority class samples, and the samples are divided into majority class and minority class. The minority class samples in I are clustered first, each sample is regarded as a separate cluster, the distance between clusters is calculated, and the two clusters with the closest distance are merged. The above process is repeated until the preset threshold T is reached. h , we get several minority class sets C = {C1, C2, ..., C n}.

[0089]

[0090] Among them, x and y are clusters C a and Cn The sample points in |C a | and |C b | represents the total number of samples in the cluster.

[0091] Furthermore, the target minority cluster can be determined according to the Euclidean distance between each initial minority cluster. Specifically, the Euclidean distance between the closest initial minority clusters is greater than the threshold T h , stop merging the initial minority clusters and get the target minority cluster Set the distance threshold T h , to determine whether to merge clusters, first define:

[0092]

[0093] Where: x p and q Clusters The sample points in Indicates the total number of samples in the cluster. avg It represents the sum of the median distances between two samples in an initial minority cluster. Taking the average of these median distances and selecting the median distance of sample distances can reduce the interference of noise samples. Define the threshold: T h =d avg f, parameter f is used to adjust the output of the clustering algorithm.

[0094] This embodiment can also determine the sampling weights corresponding to each target minority cluster. Specifically, several minority clusters with different numbers of samples can be obtained through clustering. The density within the cluster is also different, and the impact of intra-class imbalance on classification performance needs to be considered. Therefore, different weights are assigned to all minority clusters according to the number of samples, which can not only ensure that all minority clusters are oversampled and isolated small clusters are not ignored, but also help to avoid overfitting. Therefore, different sampling weights are assigned to them according to the number of samples in the minority cluster: W(i) = 1-num(i) / (∑num(i)), N is the total number of target minority clusters, and num(i) is the number of samples contained in the i-th target minority cluster. The more samples there are in a target minority cluster, the greater the proportion of the total number of samples in the target minority cluster, and the smaller the obtained W(i), that is, the smaller the assigned weight, the smaller the number of synthetic samples, and finally a balanced distribution between samples of the same type is achieved.

[0095] This embodiment can also determine the target feature data according to the sampling weight. In the classification task, the minority class samples that are closer to the decision boundary are more likely to be misclassified, which increases the difficulty of learning the minority class samples. For this reason, it is also necessary to screen the oversampled samples. The probability distribution of the minority class cluster is introduced here. According to the probability distribution, the minority class samples that are difficult to learn and contain important information are selected as seed samples to ensure the synthesis quality of the samples. The probability of the minority class cluster being selected is set as:

[0096]

[0097] Among them, P is the probability of each sample being selected, y a is the ath majority class neighbor sample of x, 1≤a≤k; is the Euclidean distance d between sample x in the minority class subcluster and sample y in the multi-class subcluster; i is a sample in the minority class cluster; n is the number of samples in a minority class cluster; k is the number of neighboring samples. The probability of each sample being selected is determined by the distance between the sample and the majority class boundary. The probability of minority class samples that are closer to the majority class boundary being selected is greater than that of samples that are farther away. The probability distribution of the minority class cluster is then formed by the probability of each sample being selected. This not only takes into account the distribution characteristics of the samples, but also effectively expands the minority class decision boundary. Therefore, feature data with a larger probability can be used as target feature data.

[0098] Step S103: Divide the target feature data to obtain a training set and a test set.

[0099] It is understandable that using the RUA-SMOTE model to generate new data for samples will make subsequent prediction results more accurate. Therefore, after resampling the samples, the sample size of the obtained data set is shown in Table 2.

[0100] Table 2

[0101] category Frequency (before sampling) Percentage (before sampling) Frequency (after sampling) Percentage (after sampling) 1 1524 4.14％ 15356 32.3％ 0 35244 95.86 32178 67.7％

[0102] It should be understood that the target feature data may be divided to obtain a training set and a test set, the training set may be 75% of the data in the target feature data, and the test set may be 25% of the data in the target feature data.

[0103] Step S104: training multiple initial gradient boosting models respectively using the training set and the test set to obtain multiple target gradient boosting models.

[0104] It is understandable that multiple initial gradient boosting models can be trained separately through training sets and test sets. This embodiment uses the initial gradient boosting models as LightGBM, XGBoost, and GradientBoosting for illustration. By establishing the LightGBM model, first of all, in order to speed up the convergence, a higher learning rate is selected, preferably around 0.1. leaming_rate, boosting_type and n_estimators are adjusted. Then adjust the decision tree, set the depth of the tree for maxdepth and num leaves, and prevent overfitting. At the same time, the GridSearchCV function in sklearn is introduced for grid search. Table 3 is a parameter table for the trained LightGBM model.

[0105] Table 3

[0106] Parameter name Numerical settings learming_rate 0.02 colsample_bylevel 0.7 colsample_bytree 0.7 Lambda_l2 10 … …

[0107] Similarly, just like building a LightGBM model, after the XGBoost model is built, some parameters of the model need to be adjusted to make the model more effective. Similarly, a higher learning rate (learingrate) is selected first, and the GridSearchCV function in sklearn is introduced to perform grid search to determine the number of trees. Then, given the learning rate and the number of trees, the decision tree is tuned, followed by regularization tuning, and finally the learning rate is reduced to determine the final parameters. Table 4 is the parameter table of the trained XGBoost model.

[0108] Table 4

[0109] Parameter name Numerical settings learming_rate 0.04 Gamma 1 N_estimator 212 Max_depth 5 … …

[0110] In the specific implementation, the grid search method is used to find the ideal number of decision trees. The most notable feature of CatBoost is that it can automatically perform one-hot encoding operations on specific attributes. At the same time, the CatBoost model can obtain higher model quality without parameter adjustment. Therefore, after several simple adjustments to the model run, the model parameters are finally determined. Table 5 is the parameter table of the trained XGBoost model.

[0111] Table 5

[0112] Parameter name Numerical settings learming_rate 0.03 l2_leaf_reg 3 Bagging_temperature 1 depth 5 … …

[0113] This embodiment performs feature processing on the initial feature data corresponding to the low-income population to obtain processed feature data, then resamples the processed feature data through a preset sampling algorithm to obtain target feature data, then divides the target feature data to obtain a training set and a test set, and then trains multiple initial gradient boosting models respectively through the training set and the test set to obtain multiple target gradient boosting models. This embodiment resamples the processed feature data through a preset sampling algorithm to obtain target feature data, which can avoid the problem of unbalanced ratio of feature data, and then accurately and effectively trains multiple initial gradient boosting models respectively through training sets and test sets.

[0114] refer to Figure 8 , Figure 8 This is a flow chart of the third embodiment of the method for identifying low-income people in the present application.

[0115] Based on the above embodiments, in this embodiment, step S20 includes:

[0116] Step S201: Determine an initial hyperparameter set corresponding to the multiple target gradient boosting models.

[0117] It is understandable that the voting model fusion method uses multiple basic machine learning models for classification problems, adopts a voting strategy, and selects the category with the most votes as the final classification. The voting fusion method is simple and effective. It only needs to use the predicted results based on the established model, and does not need to train the model again. Fig. 9 , Fig. 9 This is a diagram showing the effect of a voting algorithm according to an embodiment of a method for identifying low-income people in this application.

[0118] It should be understood that soft voting adopts the weighted average fusion method, and the weighted object is the probability of a certain model predicting a certain class. The voting parameter in the VotingClassifier in sklearn is set to sof. The formula is:

[0119] M voting =n t M1+n2M2+…+n k M k Among them, n1+n2+…+n k =1

[0120] Among them, M is the probability of each model predicting a certain category, and n is the specific weighting parameter of each model when merging, which can be assigned to each model through the weights parameter. Next, the self-built heuristic model weight adaptive algorithm is used to calculate the weight and determine H(n l ,n2…nk ) is the objective function, H(n l ,n2…n k )=-(y log(p)+(1-y)log(1-p)), the true label is y, the predicted probability given by the model is p, and log represents the natural logarithm. When y=1, the first term ylog(p) takes effect, indicating the logarithm of the probability that the model predicts a positive example when the actual example is a positive example; when y=0, the second term (1-y)l0g(1-p) takes effect, indicating the logarithm of the probability that the model predicts a negative example when the actual example is a negative example.

[0121] Randomly select initial weights Calculate H(n l ,n2…n k ) Gradient at N0 Update the current point in the direction of the gradient, calculate the tolerance, and determine whether to accept the new solution.

[0122] By starting the search at multiple random initial points at the same time, the probability of finding the global optimal solution is increased. Each initial point represents a possible model parameter configuration. The algorithm performs local searches at these points and gradually improves the parameter configuration to improve model performance. When the search at an initial point reaches the local optimum, the algorithm records the result of that point and continues to explore other initial points, using the calculated gradient ΔH multiplied by the learning rate σ as the direction of data point update, that is, For functions with boundary constraints, when a large gradient value appears, it will cause an out-of-bounds problem. The following method is used to improve it: max =(x M1 ,x M2 ,…,x Mn ),X min =(x m1 ,x m2 ,…,x mn ) are the upper and lower limits of each variable, and D is the update direction, which is as follows:

[0123]

[0124] Then, N1 = N0 + D(i) The above processing will not cause the data point update to cross the boundary. At the same time, the update rate can also be controlled according to the size of the gradient value. When the data point is already very close to the local optimal solution, updating again often results in worse results. It is necessary to timely lower the learning rate σ until the result is in the upward direction, gradually approaching the local peak.

[0125] Step S202: Optimizing the initial hyperparameter set to obtain a target hyperparameter combination.

[0126] Furthermore, in order to effectively optimize the initial hyperparameter set, in this embodiment, the step S202 includes: randomly selecting initial hyperparameter combinations corresponding to the multiple target gradient boosting models from the initial hyperparameter set; determining the expected value of the number of fitness evaluations corresponding to each of the initial hyperparameter combinations through a multi-starting point hill climbing algorithm; and selecting a target hyperparameter combination from the initial hyperparameter combinations according to the expected value.

[0127] It is understandable that there are m types of combinations in the multi-starting strategy. Each time the combination strategy is randomly selected, that is, the initial hyperparameter combinations corresponding to multiple target gradient boosting models are randomly selected from the initial hyperparameter set. The probability of each combination type being drawn is the same, and the expected number of trials required for collection is , and the probability of each peak being reached in a single execution is equal. Therefore, the peak value can be equated with the combination type, and the multi-starting traversal expectation is: E[H] = m·P m = O(mlog m), let (n, d) be a random variable, denoted as the distance between the starting weight point of n and the peak value d, and the number of fitness evaluations required to find the optimal value on a single peak problem instance. H min The expected value of (n,d), that is, the expected value of the number of fitness evaluations, is:

[0128]

[0129] It should be understood that the initial hyperparameter combination with the largest expected value can be used as the target hyperparameter combination. The self-built heuristic model weight adaptive algorithm in this embodiment significantly reduces the risk of falling into the local optimum and increases the possibility of finding the global optimal solution, thereby enhancing the prediction performance and stability of the voting model. By integrating the results of multiple searches, the multi-starting point hill climbing algorithm can provide a more robust parameter optimization solution than a single initial point, making the voting model perform better when processing complex data sets.

[0130] Step S203: Integrate the multiple target gradient boosting models according to the target hyperparameter combination to obtain a target recognition model.

[0131] It is understandable that in the process of iterative training, the three initial gradient boosting algorithms, LightGBM, XGBoost and GradientBoosting, included in the algorithm voting set, are trained by cross-validation. During the cross-validation process, different parameters are adjusted and optimized for each algorithm, including the adjustment of hyperparameters such as learning rate, tree depth, and number of leaf nodes. The optimal parameter combination is found through grid search combined with evaluation indicators (such as accuracy, recall, etc.) to improve the performance and generalization ability of the model and obtain the target gradient boosting model. The training process aims to find the best parameter configuration so that each algorithm can play its best role in the voting set, and ultimately improve the prediction accuracy and stability of the overall model. The fusion model uses the training set to train the fusion model, and the trained fusion model predicts the test set data. Finally, the prediction is compared with the label of the original data set, and the confusion matrix is ​​constructed to calculate the accuracy, recall, F1 value and AUC value to obtain the target hyperparameter combination corresponding to the target recognition model.

[0132] In the specific implementation, the trained target recognition model is uploaded to the Hadoop Distributed File System (HDFS) for deployment in the Hadoop cluster. Hadoop Streaming is used to send the data to be recognized to the Hadoop cluster, and the map task is used to predict the data to be recognized. The map task uses the trained model to predict the data to be recognized and outputs the prediction results to the standard output. The batch import / export function imports the data in batches that conform to the XMI format into the database, or exports the specified part of the database as an XMI document. The system logs the import / export process. The XMI document of the data describes the data content, attributes, and marks for adding, modifying, and deleting actions. Users can select batch import / export metadata by executing batch processing commands or screen menus, and support instant call or background timing automatic execution; use XMI to develop data templates to convert the data information in the original Word and Excel files into XMI files with fixed formats; import the XMI files with fixed formats into the data management platform in batches. Table 6 shows the output data fields of the target recognition model. It can be judged from the output data fields whether the person corresponding to the data to be recognized is a low-income person.

[0133] Table 6

[0134] Serial number Fields Field Name 1 MSISDN Number 2 USER_Type User attributes (labor force / retirees) 3 CITY_Type Place of origin (rural / urban) 4 LOW_Income Low-income people (1 yes 0 no)

[0135] This embodiment determines the initial hyperparameter set corresponding to multiple target gradient boosting models, then optimizes the initial hyperparameter set to obtain a target hyperparameter combination, and then integrates multiple target gradient boosting models according to the target hyperparameter combination to obtain a target recognition model. This embodiment optimizes the initial hyperparameter set to obtain an accurate target hyperparameter combination, and then integrates multiple target gradient boosting models according to the target hyperparameter combination. During the integration process, the weight of each target gradient boosting model is dynamically adjusted to obtain an accurate target recognition model.

[0136] Reference Fig.10 , Fig.10 This is a structural block diagram of the first embodiment of the low-income people identification device of the present application.

[0137] like Fig.10 As shown, the low-income people identification device proposed in the embodiment of the present application includes:

[0138] A model training module 10 is used to train multiple initial gradient boosting models respectively using target feature data of low-income people to obtain multiple target gradient boosting models;

[0139] A model integration module 20, for integrating the multiple target gradient boosting models to obtain a target recognition model;

[0140] The identification module 30 is used to input the data to be identified into the target identification model to obtain the low-income population identification result corresponding to the data to be identified.

[0141] In this embodiment, multiple initial gradient boosting models are trained separately through target feature data of low-income people to obtain multiple target gradient boosting models; the multiple target gradient boosting models are integrated to obtain a target recognition model; the data to be recognized is input into the target recognition model to obtain the low-income people recognition result corresponding to the data to be recognized. In this embodiment, multiple initial gradient boosting models are trained separately first, and multiple target gradient boosting models are integrated, and the weights of each target gradient boosting model are dynamically adjusted during the integration process, and then low-income people are recognized for the data to be recognized according to the target recognition model, so that the accuracy of low-income people recognition can be improved by integrating multiple gradient boosting models.

[0142] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present application. In practical applications, technicians in this field can select part or all of it according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0143] In addition, for technical details not described in detail in this embodiment, please refer to the method for identifying low-income people provided in any embodiment of the present application, and will not be repeated here.

[0144] Based on the first embodiment of the identification device for low-income people of the present application, a second embodiment of the identification device for low-income people of the present application is proposed.

[0145] In this embodiment, the model training module 10 is also used to perform feature processing on the initial feature data corresponding to the low-income population to obtain processed feature data; resample the processed feature data through a preset sampling algorithm to obtain target feature data; divide the target feature data to obtain a training set and a test set; and train multiple initial gradient boosting models separately through the training set and the test set to obtain multiple target gradient boosting models.

[0146] Furthermore, the model training module 10 is also used to screen the initial feature data corresponding to the low-income population according to the classification label to obtain the screened feature data; perform feature binning processing on the screened feature data to obtain binned feature data; perform feature cross-processing on the binned feature data to obtain processed feature data.

[0147] Furthermore, the model training module 10 is also used to eliminate noise samples in the processed feature data to obtain a sample set; select multiple minority class samples from the sample set, and cluster the multiple minority class samples to obtain initial minority class clusters; determine the target minority class cluster according to the Euclidean distance between each of the initial minority class clusters, and determine the sampling weight corresponding to each target minority class cluster; determine the target feature data according to the sampling weight.

[0148] Furthermore, the model integration module 20 is also used to determine an initial hyperparameter set corresponding to the multiple target gradient boosting models; optimize the initial hyperparameter set to obtain a target hyperparameter combination; and integrate the multiple target gradient boosting models according to the target hyperparameter combination to obtain a target recognition model.

[0149] Furthermore, the model integration module 20 is also used to randomly select initial hyperparameter combinations corresponding to the multiple target gradient boosting models from the initial hyperparameter set; determine the expected value of the number of fitness evaluations corresponding to each of the initial hyperparameter combinations through a multi-starting point hill climbing algorithm; and select a target hyperparameter combination from the initial hyperparameter combination according to the expected value.

[0150] Other embodiments or specific implementations of the low-income people identification device of the present application can refer to the above-mentioned method embodiments and will not be repeated here.

[0151] The present application provides a device for identifying low-income people, and the device for identifying low-income people includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for identifying low-income people in the above-mentioned embodiment one.

[0152] Reference below Fig.11 , which shows a schematic diagram of the structure of a low-income group identification device suitable for implementing the embodiment of the present application. The low-income group identification device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig.11 The identification device for low-income people shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0153] like Fig.11As shown, the low-income people identification device may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the low-income people identification device are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the identification device for low-income people to communicate wirelessly or wired with other devices to exchange data. Although the figure shows the identification device for low-income people with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.

[0154] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0155] The low-income people identification device provided by the present application adopts the low-income people identification method in the above embodiment, which can solve the technical problem of how to accurately and effectively identify low-income people. Compared with the prior art, the beneficial effects of the low-income people identification device provided by the present application are the same as the beneficial effects of the low-income people identification method provided by the above embodiment, and the other technical features of the low-income people identification device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0156] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0157] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0158] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the method for identifying low-income people in the above-mentioned embodiment.

[0159] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0160] The computer-readable storage medium may be included in the identification device for low-income people; or may exist independently without being assembled into the identification device for low-income people.

[0161] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the low-income people's identification device, the low-income people's identification device enables the following: to train multiple initial gradient boosting models respectively through the target feature data of the low-income people to obtain multiple target gradient boosting models; to integrate the multiple target gradient boosting models to obtain a target recognition model; to input the data to be identified into the target recognition model to obtain the low-income people identification result corresponding to the data to be identified.

[0162] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0163] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0164] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0165] The readable storage medium provided in this application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned method for identifying low-income people, and can solve the technical problem of how to accurately and effectively identify low-income people. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the method for identifying low-income people provided in the above-mentioned embodiment, and will not be repeated here.

[0166] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned method for identifying low-income people when executed by a processor.

[0167] The computer program product provided by this application can solve the technical problem of how to accurately and effectively identify low-income people. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the low-income people identification method provided by the above embodiment, which will not be repeated here.

[0168] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for identifying low-income people, characterized in that: The method for identifying low-income people comprises the following steps: Using the target feature data of low-income people, multiple initial gradient boosting models are trained separately to obtain multiple target gradient boosting models; Integrating the multiple target gradient boosting models to obtain a target recognition model; The data to be identified is input into the target identification model to obtain the low-income population identification result corresponding to the data to be identified.

2. The method for identifying low-income people according to claim 1, characterized in that: The step of training multiple initial gradient boosting models respectively using target feature data of low-income people to obtain multiple target gradient boosting models specifically includes: Performing feature processing on initial feature data corresponding to low-income people to obtain processed feature data; Resampling the processed feature data by a preset sampling algorithm to obtain target feature data; Dividing the target feature data to obtain a training set and a test set; A plurality of initial gradient boosting models are trained respectively by using the training set and the test set to obtain a plurality of target gradient boosting models.

3. The method for identifying low-income people as claimed in claim 2, characterized in that: The step of performing feature processing on the initial feature data corresponding to the low-income population to obtain the processed feature data specifically includes: The initial characteristic data corresponding to the low-income population is screened according to the classification labels to obtain the screened characteristic data; Performing feature binning processing on the filtered feature data to obtain binned feature data; Perform feature cross processing on the binned feature data to obtain processed feature data.

4. The method for identifying low-income people according to claim 2, characterized in that: The step of resampling the processed feature data by a preset sampling algorithm to obtain target feature data specifically includes: Eliminating noise samples in the processed feature data to obtain a sample set; Selecting a plurality of minority class samples from the sample set, and clustering the plurality of minority class samples to obtain an initial minority class cluster; Determine a target minority cluster according to the Euclidean distance between each of the initial minority clusters, and determine a sampling weight corresponding to each of the target minority clusters; Target feature data is determined according to the sampling weights.

5. The method for identifying low-income people according to any one of claims 1 to 4, characterized in that: The step of integrating the multiple target gradient boosting models to obtain a target recognition model specifically includes: Determining an initial set of hyperparameters corresponding to the multiple target gradient boosting models; Optimizing the initial hyperparameter set to obtain a target hyperparameter combination; The multiple target gradient boosting models are integrated according to the target hyperparameter combination to obtain a target recognition model.

6. The method for identifying low-income people according to claim 5, characterized in that: The step of optimizing the initial hyperparameter set to obtain a target hyperparameter combination specifically includes: Randomly selecting initial hyperparameter combinations corresponding to the multiple target gradient boosting models from the initial hyperparameter set; Determine the expected value of the number of fitness evaluations corresponding to each of the initial hyperparameter combinations by a multi-starting point hill climbing algorithm; A target hyperparameter combination is selected from the initial hyperparameter combinations according to the expected value.

7. A low-income people identification device, characterized in that: The low-income people identification device comprises: A model training module, used to train multiple initial gradient boosting models respectively using target feature data of low-income people to obtain multiple target gradient boosting models; A model integration module, used to integrate the multiple target gradient boosting models to obtain a target recognition model; The identification module is used to input the data to be identified into the target identification model to obtain the low-income population identification results corresponding to the data to be identified.

8. A low-income identification device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the method for identifying low-income people as described in any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the method for identifying low-income people as described in any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the method for identifying low-income people as claimed in any one of claims 1 to 6 are implemented.

Images