Vehicle Inspection Forecasting Through Temporal Signal Processing and Ensemble Machine Learning

Abstract

Vehicle inspection services are essential for maintaining transportation safety and regulatory compliance. However, daily inspection volumes in developing regions exhibit high stochastic fluctuations due to operational closures and reporting inconsistencies, making direct daily forecasting unreliable. This study proposes a weekly vehicle inspection forecasting framework using ensemble machine learning with temporal signal processing. Accurate forecasting is critical for resource allocation, staffing, and service management. The dataset covers January 2020 to December 2024 from Malang Regency, Indonesia, aggregated into 258 weekly observations. Weekly aggregation acts as a low-pass filter (ω_c = 1/7 day⁻¹) to reduce high-frequency noise while preserving seasonal dynamics. Four regression models were evaluated: Random Forest (R² = 0.6198, MAE = 69.85), XGBoost (R² = 0.7198, MAE = 55.68), Gradient Boosting (R² = 0.7373, MAE = 46.41), and a weighted ensemble (R² = 0.7164, MAE = 55.89). Conventional regression methods including Linear Regression, Ridge, and Lasso were also tested as baselines. Gradient Boosting achieved the best performance. The findings indicate that tree-based ensemble models capture non-linear temporal dynamics more effectively than conventional approaches. This study demonstrates that ensemble machine learning with signal processing provides practical forecasting tools for transportation service planning.