Smart Features, Smarter Insights: The Role of Feature Engineering in Customer Purchase Predictions in E-Commerce

Accurately predicting customer purchase intent is vital for achieving success in the e-commerce environment. While machine learning provides advanced solutions, the effectiveness of predictive models largely depends on the quality of the selected features. This study explores the impact of three feature selection methods—Information Gain, Chi-Square, and Fisher Discriminant Ratio—by utilizing real-world clickstream data to predict purchasing behavior. Our analysis reveals that the choice of selection method significantly affects the identification of key features. In particular, the Chi-Square method proved to be the most effective in minimizing class overlap, allowing for clearer differentiation between sessions that ended in a purchase and those that did not. Additionally, features associated with historical purchase probability and temporal attributes consistently demonstrated high discriminatory power. This study underscores the importance of effective feature engineering as a fundamental strategy for generating insightful predictions and enhancing model performance in understanding online consumer behavior.