IMPROVING SALES OF MAJOR COMPONENTS OF THE BIG MACHINE REMANUFACTURING SECTOR USING MACHINE LEARNING APPROACH

Abstract

This thesis research to improve the accuracy of demand predictions for remanufactured major components in Komatsu heavy equipment through a machine learning approach based on historical data. The prediction model was built using two main algorithms, Random Forest and Logistic Regression, taking into account component condition data from the Vehicle Health Monitoring System (VHMS), component replacement history, oil analysis results and sales transaction data. The research method involved DMAIC methodology, customer segmentation, classification model development, performance evaluation, and validation through business implementation. The results showed that the Random Forest algorithm provided the highest accuracy of 94%, compared to 78% for Logistic Regression. The model also successfully established a health score threshold of 30% for component condition classification.During the implementation phase, of the 157 demand opportunities generated by the model, 95 were successfully converted into actual transactions, with a conversion rate of 60% and a sales value of IDR 59 billion. These findings demonstrate that the model is not only technically valid but also has a significant impact on the effectiveness of prediction-based sales strategies and spare parts planning.