UNLOCKING REMANUFACTURING COMPONENT DEMAND POTENTIAL THROUGH BIG DATA ANALYTICS

Abstract

This study aimed to unlock remanufacturing component demand potential at United Tractors (UT) through big data analytics, specifically to predict demand for Komatsu and Scania remanufactured parts and improve demand visibility. A two-stage XGBoost model (classification and regression) was developed using comprehensive internal data (customer transactions, machine operations) and external data (weather, coal price, calendar). Model performance was evaluated using RMSE, MAE, and accuracy, comparing XGBoost against Naïve Bayes. Post-processing integrated RFM (Recency, Frequency, Monetary) analysis to prioritize sales opportunities. XGBoost consistently outperformed Naïve Bayes (RMSE 83.672 vs. 111.098; MAE 34.961 vs. 51.429). The model achieved high accuracy in 2024 (up to 98.04%) with average 70,44% but declined to ~44.5% in S1 2025, likely due to uncaptured external market shifts like sharp coal price drops. Key predictors included historical demand, agroclimate factors, and operational/commercial metrics. Adoption of analytical opportunities by UT's team significantly increased to 45-48% by May-June 2025. The research successfully applied big data analytics for remanufactured component demand prediction, with future improvements requiring dynamic adaptation to external market variables.