PREDICTIVE MAINTENANCE OF HEAVY-DUTY TRUCKS USING EXPLAINABLE MACHINE LEARNING

Abstract

Predictive maintenance for heavy-duty trucks is difficult because public failure datasets are rare and accurate machine-learning models are often hard to interpret. This study evaluates a Random Forest classifier with SHAP explanations on the SCANIA Component X dataset, a real-world multivariate time-series dataset collected from more than 33,000 trucks. The objective is to predict failure risk from the last available vehicle readout and to identify the operational variables that drive model decisions. The data were aggregated to one record per vehicle, missing values were imputed by training-set medians, and class imbalance was addressed by cost-sensitive learning and threshold optimization. On the validation set, the model achieved 0.853 accuracy, 0.447 recall, 0.315 precision and 0.370 F1-score for the failure class. SHAP analysis showed that a small set of histogram and counter variables carried most of the predictive signal.