Customer Online Purchase Behavior Prediction and Performance Analysis Using Decision Tree and Random Forest

Abstract

In the era of online shopping, understanding customer behavior has become increasingly crucial. By analyzing the detailed factors that influence customer actions, businesses can gain deeper insights into their clientele and enhance their targeted marketing strategies. This study investigates the influence of various factors on online customer purchasing frequency using a comprehensive dataset. The research is divided into two phases. Initially, the decision tree and random forest algorithms were utilized to analyze all dataset features, establishing a baseline model and determining feature importance. Subsequently, the second phase delved into the impact of feature count on model efficacy by incrementally eliminating less significant features. The study revealed that a model incorporating half of the features—namely purchase amount, age, review rating, previous purchases, location, color, purchased item, and shipping type—achieved comparable performance to the full-feature model. This streamlined approach not only expedited computation time but also reduced memory usage during the training process, offering valuable insights for businesses to refine their marketing strategies and enhance customer engagement. The findings underscore the potential of data-driven methods to optimize marketing efforts in the e-commerce sector, making a fundament for the future analysis.