With the widespread application of product recommender systems, there exists significant scope to enhance predicting customers’ purchasing possibility. Then 4 supervised classification models are constructed based on the data from a financial corporation, including logistic regression model, decision tree, bagging and random forest, to anticipate which customer is more likely to adopt the proposed health insurance policy. After comparing 𝐹1 score using an undersampling train set considering unbalanced sample data, the logistic regression model ultimately represents a superior prediction accuracy for clients’ buying propensity, and therefore has the potential to improve targeted selling system for this company.