| In the era of big data,machine learning is widely used to deal with various classification and prediction problems.Insurance cross-selling prediction can be viewed as a binary classification problem,that is,customers are interested in insurance cross-selling products and not interested in insurance cross-selling products.By establishing a fast and accurate classification model to predict whether customers are interested in insurance cross-selling products,it can help insurance companies to accurately identify the insured customer group,greatly reduce the time and cost of searching for customers,and increase company profits.Therefore,applying machine learning methods to build a predictive model suitable for insurance cross-selling data has important practical significance for insurance companies.This paper applies 381,109 insurance cross-selling data on Analytics Vidhya,and uses the support vector machine model,logistic regression model and XGBoost model that are widely used in classification problems for classification prediction.At the same time,the Ada-Voting ensemble model is innovatively proposed,which is based on Naive Bayes,decision tree,and XGBoost algorithm,combining the weight update principle of the Ada Boost model and the additive ensemble principle of the Voting model.Substitute the analyzed and processed experimental data into the SVM,Logistic Regression,XGBoost and Ada-Voting integrated models,train the model and adjust the important parameters of each model.According to the model accuracy measurement indicators(mean square error,accuracy and ROC curve AUC value)and model efficiency measure(fitting time)to comprehensively compare the four classification prediction models.The experimental results show that the Ada-Voting ensemble model performs very well in model accuracy measures such as mean square error,accuracy and ROC curve AUC value compared with the other three models,and can quickly and accurately predict insurance cross-selling data.It has a wide range of applicability and excellent application prospects.Therefore,the Ada-Voting ensemble model is finally selected to predict the data,and the importance of the data set features is sorted and analyzed according to the Ada-Voting ensemble model.The results show Vehicle Damage,Previously Insured and Age are the top three variables in the order of importance.The analysis results from the Ada-Voting ensemble model will help insurance companies formulate targeted marketing plans and effectively implement cross-selling of insurance products.Therefore,the Ada-Voting ensemble model proposed in this paper provides more innovative ideas for practical application fields such as cross-selling and theoretical research on classification problems. |
