Bank marketing customer screening method based on cross-algorithm feature screening

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a bank marketing customer screening method based on cross-algorithm feature screening, and the method comprises the steps: firstly obtaining multi-dimensional data of customers, and carrying out the preprocessing and feature engineering; screening the data after the feature engineering by adopting cross-algorithm fusion feature importance evaluation to obtain a feature set; in combination with the screened feature set, performing model training based on an automatic machine learning framework; deploying the trained model into a screening system, wherein the prediction result of the model is used for guiding the adjustment of a marketing strategy; according to the invention, a cross-algorithm fusion feature screening method and an automatic machine learning technology are utilized to effectively solve the problems of efficiency and precision in accurate screening of bank customers; through dynamic weight adjustment, a self-adaptive layering strategy and a real-time optimization mechanism, the method has relatively high technical innovation and business adaptability, and can be widely applied to the field of financial marketing.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and particularly to a method for screening bank marketing customers based on cross-algorithm feature screening. Background Art

[0002] Bank marketing activities are comprehensive activities carried out by banks to meet customer needs and enhance market competitiveness, covering aspects such as market research, product development, brand promotion, and customer relationship maintenance. Its essence is to establish a long-term and stable cooperative relationship with customers by creating and delivering value. In traditional bank marketing activities, accurately screening target customers has always been a difficult problem.

[0003] For example, in the prior art, a single algorithm is usually relied on for feature selection and model construction. However, in complex data scenarios, a single algorithm is difficult to fully exploit the potential value of input features. At the same time, traditional model training requires a large amount of manual intervention, resulting in low efficiency. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for screening bank marketing customers based on cross-algorithm feature screening, aiming to solve the technical problems in the prior art that usually rely on a single algorithm for feature selection and model construction, but in complex data scenarios, a single algorithm is difficult to fully exploit the potential value of input features, and at the same time, traditional model training requires a large amount of manual intervention, resulting in low efficiency.

[0005] To achieve the above purpose, a method for screening bank marketing customers based on cross-algorithm feature screening adopted by the present invention includes the following steps:

[0006] Firstly, obtain multi-dimensional data of customers and perform preprocessing and feature engineering;

[0007] Use cross-algorithm fusion-based feature importance evaluation to screen the data after feature engineering to obtain a feature set;

[0008] Combine the screened feature set and perform model training based on an automated machine learning framework;

[0009] Deploy the trained model into a screening system, and the prediction results of the model are used to guide the adjustment of marketing strategies;

[0010] During the application process, the system automatically decides whether to update the model version in the current production environment.

[0011] Among them, the multi-dimensional data of customers obtained includes personal basic information, historical transaction records, and asset-liability situations;

[0012] Preprocessing the data and performing feature engineering include feature processing and feature generation. Feature processing includes handling missing values, outliers, binning continuous features, and normalization;

[0013] Feature generation includes expanding the feature space by aggregating statistical features, time series features, and cross features.

[0014] Among them, multiple algorithms are combined for evaluation to strengthen the robustness of feature screening; the specific evaluation formula is as follows:

[0015] Comprehensive feature importance =

[0016] w1 * Information Gain + w2 * IV value + w3 * model1 + …… + wn * modelN;

[0017] Among them, the logic for determining the weights of w1, w2, and w3 is as follows:

[0018] At the same time, calculate the information gain, IV value, and feature importance based on tree models (such as XGBoost or LightGBM, etc.) for each feature. Each feature importance evaluation method ranks all features and selects the top N features;

[0019] After obtaining the three sequences of TOP N, we initialize the weights of the three types to 1 / 3, train models for each feature subset respectively and record their AUC values, and adjust the weights based on the model performance;

[0020]

[0021] Using the adjusted weights w1, w2, and w3, calculate the comprehensive importance of each feature:

[0022] Comprehensive feature importance = w1 * Information Gain + w2 * IV value + w3 * model;

[0023] Normalize the comprehensive importance of all features:

[0024]

[0025] Set the feature selection threshold, select features with comprehensive importance greater than a certain value, or directly select the top m features as the final initial feature selection, and then perform subsequent model training.

[0026] Among them, based on the selected feature set, the specific method for model training based on the automated machine learning framework is as follows:

[0027] Build a model search space through the AutoML framework, covering multiple algorithms (such as XGBoost, LightGBM, RandomForest, etc.), and automatically select a single best model or the optimal fused model according to performance metrics (such as AUC, F1 score, R2, etc.);

[0028] At the same time, the AutoML framework incorporates efficient hyperparameter tuning algorithms (such as Bayesian optimization), which can predict the optimal solution in the parameter space based on historical evaluation results.

[0029] Among them, the specific way to use the prediction results of the model to guide the adjustment of marketing strategies is as follows:

[0030] Link the model prediction results with predefined marketing strategies (such as personalized offers, precise push, etc.) to form a rule-driven automatic execution framework;

[0031] At the same time, by collecting user feedback on the strategies, the prediction model and strategy rules are optimized in real time to improve marketing efficiency and effectiveness.

[0032] Among them, the ways for the system to automatically determine whether to update the model version in the current production environment include the model comparison strategy and the progressive rollout strategy.

[0033] Among them, the model comparison strategy is as follows: Before the new model is launched, it needs to be compared with the current production model, and it is judged whether to replace it through key metrics (such as AUC, F1 score, etc.).

[0034] Among them, the progressive rollout strategy is as follows: In the production environment, the system will first apply the new model to some samples, verify the superiority of the new model through the actual effect, and then gradually expand the application scope;

[0035] If the effect of the new model is not as expected after it is launched, the system will automatically roll back to the old version model to ensure business continuity.

[0036] A bank marketing customer screening method based on cross-algorithm feature screening of the present invention, in specific use, first obtains multi-dimensional data of customers, and performs preprocessing and feature engineering; uses cross-algorithm fusion feature importance evaluation to screen the data after feature engineering to obtain a feature set; combines the screened feature set, and performs model training based on an automated machine learning framework; deploys the trained model into the screening system, and the prediction results of the model are used to guide the adjustment of marketing strategies; during the application process, the system automatically determines whether to update the model version in the current production environment. The present invention uses the cross-algorithm fusion feature screening method and automated machine learning technology to effectively solve the efficiency and accuracy problems in bank customer precise screening. Through dynamic weight adjustment, adaptive hierarchical strategy and real-time optimization mechanism, it has strong technical innovation and business adaptability, and can be widely applied to the financial marketing field. Description of the Drawings

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0038] Figure 1 It is a flowchart of the bank marketing customer screening method based on cross-algorithm feature screening of the present invention. Detailed Embodiments

[0039] The following will describe in detail the embodiments of the present invention. The examples of the embodiments are shown in the drawings. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present invention and should not be construed as a limitation to the present invention.

[0040] Please refer to Figure 1 , Figure 1 It is a flowchart of the bank marketing customer screening method based on cross-algorithm feature screening of the present invention.

[0041] The present invention provides a bank marketing customer screening method based on cross-algorithm feature screening, including the following steps:

[0042] S1. First, obtain the multi-dimensional data of the customer and perform preprocessing and feature engineering;

[0043] For this specific embodiment, the multi-dimensional data of the customer obtained includes personal basic information, historical transaction records, and asset-liability situation;

[0044] Performing preprocessing and feature engineering on the data includes feature processing and feature generation. Feature processing includes handling missing values, outliers, binning continuous features, and normalization processing;

[0045] Feature generation includes expanding the feature space by aggregating statistical features, time series features, and cross features.

[0046] S2. Use cross-algorithm fusion feature importance evaluation to screen the data after feature engineering to obtain a feature set;

[0047] For this specific embodiment, cross-algorithm fusion feature importance evaluation is used to screen features. Traditional methods (such as information gain, IV value, etc.) may have limited performance under different data distributions. The present invention combines multiple algorithms for evaluation to enhance the robustness of feature screening. The specific evaluation formula is as follows:

[0048] Comprehensive feature importance =

[0049] w1 * Information Gain + w2 * IV value + w3 * model1 + …… + wn * modelN;

[0050] Among them, the weight determination logic of w1, w2, and w3 is as follows:

[0051] S201. At the same time, calculate the information gain, IV value, and feature importance based on the tree model (such as XGBoost or LightGBM, etc.) of each feature. Each feature importance evaluation method sorts all features and selects the top N features;

[0052] For example:

[0053] Information Gain: Calculate the information gain between each feature and the target variable, and sort to select the top N features (information_gain).

[0054] IV value: Calculate the IV value of each feature, and sort to select the top N features. (iv_value)

[0055] Tree Model Gain: Train a tree-based model (such as XGBoost), extract the feature importance, and sort to select the top N features. (tree_model_gain).

[0056] The above 3 methods can also filter the number of features according to a threshold.

[0057] S202. After obtaining the 3 types of TOP N sequences, we initialize the weights of the 3 types to 1 / 3, train models for each feature subset respectively and record their AUC values, and adjust the weights based on the model performance.

[0058]

[0059] The implementation code is as follows:

[0060]

[0061]

[0062] S203. Using the adjusted weights w1, w2, w3, calculate the comprehensive importance of each feature:

[0063] Feature comprehensive importance = w1 * Information Gain + w2 * IV value + w3 * model; Normalize the comprehensive importance of all features:

[0064]

[0065] The implementation code is as follows:

[0066]

[0067] S204. Set the feature selection threshold, select features with a comprehensive importance greater than a certain value, or directly select the top m features as the final initial feature selection, and then perform subsequent model training.

[0068] # Set the threshold or select the top m features

[0069] threshold = 0.5 # Comprehensive importance threshold

[0070] selected_features = [feature for feature, importance in final_importance.items() if importance >= threshold]

[0071] # Or select the top m features

[0072] m = 5

[0073] selected_features = list(final_importance.keys())[:m]

[0074] print("Final selected features:", selected_features)

[0075] S3. Combine the selected feature set and perform model training based on the automated machine learning framework;

[0076] For this specific embodiment, the present invention adopts an automated machine learning framework (AutoML), which significantly reduces the need for manual intervention by automatically performing model selection, hyperparameter optimization, feature engineering, etc. The beneficial effects are as follows:

[0077] Automatic model search:

[0078] The system constructs a model search space through the AutoML framework, covering a variety of algorithms (such as XGBoost, LightGBM, Random Forest, etc.), and automatically selects the optimal algorithm according to the task requirements (classification or regression).

[0079] Automatic hyperparameter tuning:

[0080] AutoML incorporates an efficient hyperparameter tuning algorithm (such as Bayesian optimization), which can predict the optimal solution in the parameter space based on historical evaluation results, avoiding the high computational cost of grid search.

[0081] Model Fusion and Selection:

[0082] Fusion Strategy: After multi-model evaluation, the system fuses multiple excellent-performing models through techniques such as weighted average or Stacking to improve the overall prediction performance.

[0083] Selection Logic: The AutoML framework automatically selects the single best model or the optimal fused model according to performance metrics (such as AUC, F1 score, R2, etc.) to adapt to different business scenarios.

[0084] S4. Deploy the trained model into the screening system, and the prediction results of the model are used to guide the adjustment of the marketing strategy;

[0085] For this specific embodiment, the specific way of using the prediction results of the model to guide the adjustment of the marketing strategy is:

[0086] Link the model prediction results with predefined marketing strategies (such as personalized offers, precise push, etc.) to form a rule-driven automatic execution framework;

[0087] Meanwhile, by collecting users' feedback on the strategy, the prediction model and strategy rules are optimized in real time to improve the marketing efficiency and effect.

[0088] Linkage between the Fusion Model and Business Strategy:

[0089] The prediction results of the model need to be efficiently integrated into the business strategy to guide marketing behaviors. The present invention realizes the linkage through the following methods:

[0090] Rule Engine Driving:

[0091] Build a rule engine to map the model outputs (such as user scores, classification results, etc.) to specific marketing strategies (such as pushing coupons, adjusting interest rates, etc.).

[0092] The engine supports flexible configuration, facilitating quick adaptation to the needs of strategy adjustment.

[0093] Online Feedback and Optimization:

[0094] Collect the real-time feedback of users on the strategy and add it as new features to the AutoML process to further optimize the model.

[0095] Form a closed loop of "model - strategy - feedback - model" to continuously improve the business effect.

[0096] S5. During the application process, the system automatically decides whether to update the model version in the current production environment.

[0097] For this specific embodiment, the ways for the system to automatically determine whether to update the model version in the current production environment include the model comparison strategy and the progressive online strategy.

[0098] Among them, the model comparison strategy is as follows: Before the new model goes online, it needs to be compared with the current production model, and it is judged whether to replace it through key indicators (such as AUC, F1 score, etc.).

[0099] Among them, the progressive online strategy is as follows: In the production environment, the system will first apply the new model to some samples, verify the superiority of the new model through the actual effect, and then gradually expand the application scope;

[0100] If the effect after the new model goes online is less than expected, the system will automatically roll back to the old version model to ensure business continuity.

[0101] When using a bank marketing customer screening method based on cross-algorithm feature screening of the present invention, in specific use, first obtain the multi-dimensional data of customers, and perform preprocessing and feature engineering; use the feature importance evaluation of cross-algorithm fusion to screen the data after feature engineering to obtain a feature set; combine the screened feature set, and perform model training based on the automated machine learning framework; deploy the trained model into the screening system, and the prediction results of the model are used to guide the adjustment of marketing strategies; during the application process, the system automatically determines whether to update the model version in the current production environment. The present invention uses the cross-algorithm fusion feature screening method and automated machine learning technology to effectively solve the efficiency and accuracy problems in the precise screening of bank customers. Through dynamic weight adjustment, adaptive stratification strategy and real-time optimization mechanism, it has strong technological innovation and business adaptability, and can be widely applied to the field of financial marketing.

[0102] Compared with the traditional manual method, the feature screening efficiency of the present invention is increased by more than 50%, and the screening result is more accurate. The present invention improves the customer conversion rate: The identification of accurate target customers increases the bank marketing conversion rate by 20%-30%. The present invention realizes rapid deployment and iteration through an automated process to cope with different business scenarios.

[0103] What is disclosed above is only a preferred embodiment of the present invention. Of course, the scope of the rights of the present invention cannot be limited by this. Those of ordinary skill in the art can understand all or part of the processes of implementing the above embodiments, and the equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A method for screening bank marketing customers based on cross-algorithm feature screening, characterized in that it includes the following steps: First, obtain the multi-dimensional data of customers, and perform preprocessing and feature engineering; Use cross-algorithm fusion feature importance evaluation to screen the data after feature engineering to obtain a feature set; Based on the selected feature set, perform model training based on an automated machine learning framework; Deploy the trained model into the screening system, and the prediction results of the model are used to guide the adjustment of marketing strategies; During the application process, the system automatically decides whether to update the model version in the current production environment.

2. The method for screening bank marketing customers based on cross-algorithm feature screening according to claim 1, characterized in that the obtained multi-dimensional data of customers include personal basic information, historical transaction records, and asset-liability situations; Performing preprocessing and feature engineering on the data includes feature processing and feature generation. Feature processing includes handling missing values, outliers, binning continuous features, and normalization processing; Feature generation includes expanding the feature space by aggregating statistical features, time series features, and cross features.

3. The method for screening bank marketing customers based on cross-algorithm feature screening according to claim 2, characterized in that The method of using cross-algorithm fusion feature importance evaluation to screen the data after feature engineering to obtain a feature set is as follows: Combine multiple algorithms for evaluation to strengthen the robustness of feature screening; the specific evaluation formula is as follows: Feature comprehensive importance = w1 * Information Gain + w2 * IV value + w3 * model1 + …… + wn * modelN; Among them, the weight determination logic of w1, w2, and w3 is as follows: Simultaneously calculate the information gain, IV value, and feature importance based on the tree model of each feature. Each feature importance evaluation method sorts all features and selects the top N features; After obtaining the three sequences of TOP N, we initialize the weights of the three types to 1 / 3, train models for each feature subset respectively and record their AUC values, and adjust the weights based on model performance; Using the adjusted weights w1, w2, and w3, calculate the comprehensive importance of each feature: Feature comprehensive importance = w1 * Information Gain + w2 * IV value + w3 * model; Normalize the comprehensive importance of all features: Set a feature selection threshold, select features with comprehensive importance greater than a certain value, or directly select the first m features as the final initial feature selection, and then perform subsequent model training.

4. The method for screening bank marketing customers based on cross-algorithm feature screening according to claim 3, characterized in that The specific method of performing model training based on an automated machine learning framework in combination with the selected feature set is as follows: Construct a model search space through the AutoML framework, covering multiple algorithms, and automatically select a single best model or the optimal model after fusion according to performance indicators; At the same time, the AutoML framework has built-in efficient hyperparameter tuning algorithms (such as Bayesian optimization), which can predict the optimal solution in the parameter space based on historical evaluation results.

5. The bank marketing customer screening method based on cross-algorithm feature screening as claimed in claim 4, wherein The specific way of using the prediction result of the model to guide the adjustment of the marketing strategy is as follows: Link the model prediction result with the predefined marketing strategy to form a rule-driven automatic execution framework; At the same time, by collecting the feedback of users on the strategy, the prediction model and the strategy rules are optimized in real time to improve the marketing efficiency and effect.

6. The bank marketing customer screening method based on cross-algorithm feature screening as claimed in claim 5, wherein The ways for the system to automatically determine whether to update the model version in the current production environment include the model comparison strategy and the progressive online strategy.

7. The bank marketing customer screening method based on cross-algorithm feature screening as claimed in claim 6, wherein The model comparison strategy is adopted as follows: Before the new model is launched, it needs to be compared with the current production model, and it is judged whether to replace it through key indicators.

8. The bank marketing customer screening method based on cross-algorithm feature screening as claimed in claim 7, wherein The progressive online strategy is adopted as follows: In the production environment, the system will first apply the new model to some samples, and verify the superiority of the new model through the actual effect, and then gradually expand the application scope; If the effect after the new model is launched is less than expected, the system will automatically roll back to the old version model to ensure business continuity.

Images