Application Research Of Deep Learning Model In Bank Customer Churn Prediction

In recent years,with the popularity of mobile devices and the rapid development of mobile Internet,the rapid growth of data has led to the rapid development of emerging industries such as Internet finance.Under this background,the traditional financial industry has been severely impacted by the Internet finance industry.The rapid growth of data is a "double-edged sword",which not only brings more management costs to enterprises,but also contains more valuable information.The problem of customer churn prediction is one of the most discussed problems in banking business.Exploring a customer churn prediction model suitable for the existing data environment is of great significance to the development of banking business.In this thesis,after analyzing and researching the historical deposit and loan,transaction statistics and credit information of bank customers,data preprocessing and feature selection are carried out to obtain the input features of the model.In the prediction experiment of 104224 users,firstly,two traditional machine learning models,CART classification tree and random forest,are used for simulation experiments.The accuracy of CART classification tree is 55.05%,and the accuracy of random forest is 63.17%,which is of low application value in actual production.In order to improve the simulation effect and explore the application of deep learning models in this data set,the accuracy of the experimental results of the four proposed deep learning models has reached more than 87%.Among them,the BLSTM-CNN model,which integrates RNNs and CNNs in parallel,solves the defect problem of the separate simulation of RNNs and CNNs,and solves the problem of DLCNN model LSTM layer output results that ignore part of the local information when input to the convolutional layer The accuracy of the experiment reached 95.33%;and in the process,a method of reconstructing two-dimensional feature data into three-dimensional feature data was proposed,which constructed the dynamic features of the two-dimensional data and increased the input features of the model;Based on the BLSTM-CNN model,the attention mechanism is introduced,and the Attn BLSTMCNN model is proposed.Compared with the BLSTM-CNN,the accuracy is increased by 0.2%,which further improves the effect of the model.Finally,according to the Attn BLSTM-CNN model,the application process of the actual production,from data processing to model training,and finally obtained prediction results.