Application Research Of User Churn Prediction Based On Machine Learning Model

With the popularity of mobile devices and mobile Internet,data is showing a rapid growth trend,and while promoting the prosperity of Internet finance,it also has a certain degree of impact on the traditional financial industry.The proliferation of bank data has led to increasing costs for its management and more valuable data information,which means new opportunities and challenges for bank operations and decision making.Bank operations and decisions are tied to user stickiness,and when banks make mistakes in their decisions,it can lead to user churn,so the user churn problem is one of the urgent issues that banks need to address.This study aims to achieve accurate prediction of bank user churn by designing a machine learning based bank user churn prediction model.Through the prediction results of the model,banks can take targeted measures to retain users who are about to churn,improve the accuracy of user management and service enhancement,and enhance market competitiveness while improving the bank’s sustainability.To address the problem of low classification and recognition accuracy of existing user churn prediction models in the banking field when dealing with a large amount of highdimensional data,this paper proposes a Long Short-Term Memory Attention(LSTM-att)prediction model that incorporates an attention mechanism and a Mask mechanism to perform global attention calculation of data through the attention layer.The computation process processes the training data by a certain proportion of Mask masks,after which the correspondence between the attention weights and the original data is obfuscated to further enhance the training effect of the model.Finally,the processed data are put into the LSTM neural network for training,and finally the prediction is performed in the Softmax layer and the model is optimized by calculating the loss.To address the problems that the single machine learning bank user churn prediction model is not significantly improved and the features in the sample are not sufficiently extracted,this study adopts a multi-model fusion strategy to make up for the shortcomings of the traditional model and obtain further performance improvement in the bank user churn prediction model by this method.Based on the LSTM-att model,the Cla-XGB(Convolutional Neural Networks-Long Short Term Memory-Attention-e Xtreme Gradient Boosting)bank user churn prediction model based on multi-model fusion is proposed.In addition,to further improve the prediction performance of the model,SMOTE oversampling and focused loss function are introduced to further optimize the impact caused by data imbalance.Using the public dataset provided by kaggle,this paper demonstrates the effectiveness of the improved prediction model and the superiority of the prediction performance proposed in this paper through ablation experiments and comparison experiments,and provides a feasible new approach for the sustainable development of the financial industry.