Comparative analysis of machine learning algorithms for failure prediction of construction and road machinery

Abstract

One of the key challenges in the operation of construction and road machinery is predicting equipment failures before they lead to unplanned downtime. This paper compares three machine learning algorithms – Random Forest, Gradient Boosting, and Support Vector Machine (SVM with RBF kernel) – for binary classification of equipment technical condition. The experiments were conducted on the open-access AI4I 2020 Predictive Maintenance Dataset (UCI ML Repository), which contains 10,000 records with sensor readings: air and process temperatures, rotational speed, torque, and tool wear. The failure rate in the dataset is 3,4%, which is representative of real monitoring systems. Models were evaluated using Accuracy, F1-Score, and ROC-AUC. The best ROC-AUC of 0.972 was achieved by Random Forest; Gradient Boosting led in F1-Score (0.768). Feature importance analysis revealed torque (0.327) and rotational speed (0.293) as the primary failure predictors. The results confirm the potential of ensemble methods for predictive maintenance systems in construction machinery.