A Comparative Study on Data Mining Algorithms for Predicting Customer Churn

This study investigates predicting customer churn in Indonesia's dynamic forestry industry through machine learning techniques.Models were developed leveraging five years of transactional data from Perum Perhutani, contrasting ten algorithms, we address the research problem by following CRISP-DM methodology.The Gradient Boosted Tree achieved the highest accuracy of 85.60% and AUC of 0.940 in discriminating churners from non-churners.Analysis of feature importance reveals transaction timing attributes as most influential, underscoring recency, frequency, and intervals as pivotal indicators of attrition.Findings provide actionable insights guiding customer retention initiatives and risk mitigation strategies for industrial suppliers in turbulent economic environments.