Predicting Online Repeat Purchases: A Comparative Analysis of Machine Learning Algorithms

Post-COVID, there has been a rise in online book shopping, increasing the need for targeted online marketing to meet shifting customer preferences. This study evaluates advanced machine learning algorithms, including Voting Classifier, CatBoost, XGBoost, Random Forest, and Linear SVM to predict repeat purchases in an online bookstore context using Amazon sales data. The research addresses challenges posed by class imbalance using the Synthetic Minority Over-sampling Technique - Edited Nearest Neighbors(SMOTE-ENN), a hybrid resampling method. The Voting Classifier based on ensemble methods emerged as the most effective model, achieving high accuracy, precision, recall, and AUC scores, demonstrating its robustness and reliability. The findings emphasize the importance of advanced algorithms, interpretability, and comprehensive data preparation in improving the purchase prediction of the target customer group. The study’s insights can help businesses enhance strategic decision-making and targeted marketing, ultimately boosting customer retention and revenue in e-commerce settings.