Enhancing Repeat Buyer Classification with Multi Feature Engineering in Logistic Regression

This study presents a novel approach to improving repeat buyer classification on e-commerce platforms by integrating Kullback-Leibler (KL) divergence with logistic regression and focused feature engineering techniques. Repeat buyers are a critical segment for driving long-term revenue and customer retention, yet identifying them accurately poses challenges due to class imbalance and the complexity of consumer behavior. This research uses KL divergence in a new way to help choose important features and evaluate the model, making it easier to understand and more effective at classifying repeat buyers, unlike traditional methods. Using a real-world dataset from Indonesian e-commerce with 1,000 records, divided into 80% for training and 20% for testing, the study uses logistic regression along with techniques like SMOTE for oversampling, class weighting, and regularization to fix issues with data imbalance and overfitting. Model performance is assessed using accuracy, precision, recall, F1-score, and KL divergence. Experimental results indicate that the KL-enhanced logistic regression model significantly outperforms the baseline, especially in balancing precision and recall for the minority class of repeat buyers. The unique contribution of this work lies in its synergistic use of KL divergence in both the feature engineering and evaluation phases, offering a robust, interpreted, and data-efficient solution. For e-commerce businesses, the findings translate into improved targeting of high-value customers, better personalization of marketing efforts, and more strategic allocation of resources. This research offers practical tips for enhancing predictive customer analytics and supports data-driven decision-making in digital commerce environments.