Optimizing E-Commerce in Indonesia: Ensemble Learning for Predicting Potential Buyers

Abstract

In the competitive Indonesian e-commerce sector, data-driven decisionmaking is crucial for success. This study addresses the challenge faced by a leading e-commerce company, where despite a 134% increase in promotional expenses, active user transactions remained low. Focusing on predicting potential buyers to optimize promotional spending, the research evaluates various ensemble learning methods, including Random Forest, XGBoost, and LightGBM algorithms. Through extensive testing, all three models demonstrated high precision in identifying potential buyers. Remarkably, XGBoost achieved an exceptional precision score of 89.5%. Further enhancement through a soft voting strategy combining XGBoost and LightGBM resulted in the highest precision rate of 89.8%, suggesting a promising approach for targeted marketing and improved promotional strategies in the e-commerce industry