Detecting high fuel consumption in HDVs with ensemble of anomaly detection models

This paper presents a machine learning (ML)-based system designed to detect high fuel consumption in heavy-duty vehicles (HDVs) using operational data. The system addresses environmental and efficiency challenges in the transportation industry by precisely monitoring fuel consumption to curb CO2 emissions. An ensemble learning method that integrates unsupervised anomaly detection techniques, including Isolation Forest, Autoencoder, and k-NN Regressor models, is proposed. The anomaly detection results from these models are combined using a weighted majority voting (WMV) approach. This method was tested on a dataset comprising 459 driving records and 14 signals collected from 187 HDVs. Additionally, the Local Outlier Factor (LOF) model was employed to validate the ensemble learning method and identify the root causes of the anomalies. This work contributes to the field of transportation efficiency by offering a novel approach to analyzing fuel consumption in HDVs, thereby paving the way for future advancements in sustainable transportation practices.