BOOSTING WARRANTY PREDICTION IN THE AUTOMOTIVE SECTOR USING MACHINE LEARNING: A COMPARATIVE STUDY

This paper investigates the application of machine learning techniques for reliability forecasting in the automotive industry, focusing on the prediction of Months In Service (MIS), that is, the number of months between a vehicle's warranty start date and repair date. The accurate estimation of MIS is of crucial importance to improve the forecast of warranty costs, reliability planning, and maintenance strategies. Traditional reliability tools, such as Weibull analysis, work effectively under strict statistical assumptions but often fail to capture nonlinear relationships and interdependencies within complex high-dimensional field data. The aim of this paper is to compare two machine learning algorithms: Decision Trees (DT) and Gradient Boosting Decision Trees (GBDT). Their performance was compared based on real industrial warranty data from a Tier-1 automotive supplier. The dataset includes categorical, numerical, temporal, and operational variables describing vehicle characteristics and usage conditions. This comparison underlines the fact that the ensemble learning approach significantly outperforms the single-tree model in accuracy, generalization capability, and robustness when predicting MIS. Experimental results confirm the superiority of the boosting algorithm. These observations allow the conclusion that ensemble methods really improve the results of reliability forecasting due to modeling complex nonlinear dependencies. This work is concluded by verifying that boosting-based models provide a flexible, accurate, and data-driven approach; it opens new frontiers to improve current methodologies for reliability analysis and warranty management in the automotive sector.

Keywords: Warranty prediction, machine learning, Weibull distribution, automotive industry, MIS, GBDT, DT.