Bank Customers Data Analysis And Prediction Based On Interpretable Machine Learning Model

In the traditional industry,the main task of machine learning model is to solve the problems in real life.It is more inclined to apply the model to data,rather than explain why the model can achieve such good results.In other specific fields,such as finance,we can't flow training data into the black box model and train a function(this function can also be called a model)like the traditional machine learning model,and input new data to the function to get the prediction results,these are not enough.We also need to analyze the interpretability of the model.The interpretability has a very positive significance for the verification and improvement of the model.How to explain the model,how it is predicted,and where the credibility of the model is reflected are crucial for the development of banking business.Only by doing this,can we really apply machine learning model to the field of banking financial innovation,let machines learn from people's experience,give managers better decision-making,realize the target marketing of the financial industry.And truly use the ability of data and machine learning model to solve practical financial problems.As the main representative of the financial field,the bank's main business is time deposit,but the rapid development of the Internet has caused a great impact on the bank's financial products.More and more people choose the bank's time deposit business to the Internet financial products,which means a large loss of customer resources for the bank.The bank's goal is precision marketing,which is to find customers with lifelong value,or customers who choose to make lifelong time deposits in the bank,and then maintain the long-term cooperative relationship between customers and the bank,so that they can choose the bank's financial products.Bank managers always hope that risk control analysts can accurately predict whether customers will subscribe the bank's products,which factors affect customers' subscription and how to improve them,which all play a crucial role in the economic benefits of banks,and we hope to determine which high-quality customers can help the bank achieve the marketing revenue exceeding the expected cost,and truly achieve the precise marketing of the bank's financial products.Based on the 21 influencing factors of whether bank customers subscribe time deposits,this paper analyzes the interpretability of the model from the following three aspects: The first is the interpretability method before establishing the model,mainly starting from the data itself,using the method of data analysis,mining the rules from the data,having a comprehensive understanding of the distribution and feature relationship of the data,and then carrying out data visualization,feature engineering,unbalanced data processing,etc.The second is the explainability of the modeling.In this paper,the Ensemble learning model and the deep learning model LSTM are combined to predict,and the Recall,Precision,Accuracy and AUC value are used as the model evaluation indexes.Because the Ensemble learning model has strong explainability,but the deep learning model has poor explainability,so we need to analyze the explainability of the deep learning model,mainly using Agent Model to explain the deep learning model.The third is the interpretable methods after building the model,which are associated with the model.Some other methods are called model independent interpretation,including Feature Importance analysis,Partial Dependence Plot,Individual Conditional Expectation,Shapley Values and so on.Through the interpretability analysis of three aspects,we can understand the model more transparently and increase the credibility of the model.This is a good explanation for whether customers subscribe time deposits,which is conducive to the bank's precise marketing according to customer characteristics and increase the bank's income.