Prediction Of Customer Deposit Behavior Based On Machine Learning Algorithms

With the rapid growth of financial technology,traditional banking industry is actively exploring intelligent transformation.Deposit business is the major liability business of banks.Predicting customer deposit behavior accurately,identifying potential and target customers and carrying out precise marketing with Machine Learning Algorithms can effectively improve customer conversion rate,which is crucial to successful transformation.This paper selects the information generated in the deposit marketing activities of a commercial bank in Portugal as the dataset,and builds Random Forest and XGBoost to predict whether customers will order deposit,which provides certain reference for bank deposit marketing in China.This paper constructs traditional Random Forest and XGBoost through parameter optimization and feature selection,and evaluates two models through feature importance and Confusion Matrix.Then,this paper improves the traditional Random Forest.The weights of all Decision Trees are the same by default in the integrated voting of traditional Random Forest.In fact,their performances are not the same.Giving each Decision Tree the certain weight can increase the influences of those excellent Decision Trees on the final results,thereby improving the Random Forest accuracy.From this perspective,this paper optimizes the Random Forest.First,those Decision Trees with low accuracy are eliminated by pruning.Second,the remaining trees with high accuracy are given corresponding weights based on the Classification Margin change,and this generates a weighted ensemble Random Forest.Later,the paper uses the joint ROC and PR curves to compare the classification performances of the three models,uses the intersection of recall rate and precision rate curves to determine the optimal range of the threshold,and uses Confusion Matrix to compare the customer classification results under the three models.The result shows that compared with traditional Random Forest and XGBoost,the improved Random Forest can more accurately predict deposit behavior,effectively improves customer classification accuracy,and has certain reference significance for banks to carry out deposit marketing activities.Finally,based on the feature importance ranking,the paper determines several important feature variables that affect customer deposit,and proposes precise marketing strategies for banks to carry out deposit telemarketing activities by analyzing the distribution rules of these important features.