Fuel consumption classification for heavy-duty vehicles: a novel approach to identifying driver behavior and system anomalies

In this paper, we propose a fuel consumption classification system for heavy-duty vehicles (HDVs) based on two machine learning models that categorize sections of driving data as normal or high and inlier or outlier fuel consumption. A dataset of 606 naturalistic driving records collected from 57 different heavy-duty trucks with varying carry loads is generated and utilized. Proposed models are trained to categorize driving sections taking into consideration of vehicle weight and road slope, which are the two major factors affecting the fuel consumption of a heavy-duty truck. Results show an accuracy of 92.2% in high fuel consumption prediction and an F1 score of 0.78 in outlier prediction using the bagged decision trees models. The proposed approach provides an advanced categorization of driving data in terms of fuel economy. It has substantial potential to determine driving behavior anomalies or system faults that may cause excessive energy consumption and emissions in HDVs.