Vehicle Fuel Optimization Under Real-World Driving Conditions: An Explainable Artificial Intelligence Approach

Barbado, Alberto; Corcho, Óscar · 2021 · OpenAlex-citations

DOI: 10.48550/arxiv.2107.06031

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical need for reducing fuel consumption and associated emissions in industrial vehicle fleets, particularly diesel and petrol vehicles. While transitioning to electric vehicles is a long-term solution, the authors argue that immediate cost and emission reductions are necessary given the slow adoption rate of electric vehicles. The research focuses on optimizing driving behavior and other actionable factors to lower fuel usage. To achieve this, the paper introduces an Explainable Artificial Intelligence (XAI) approach using the Explainable Boosting Machine (EBM) algorithm. Unlike traditional black-box machine learning models that lack interpretability or linear models that cannot capture complex non-linear relationships, EBM provides transparent, feature-relevance-based explanations that quantify the individual impact of specific input factors on fuel consumption. The methodology utilizes real-world telematic data collected from over 1,000 industrial vehicles between 2020 and 2021. The dataset includes up to 70 features derived from On-Board Diagnostics (OBD-II) and other sources, categorized into groups such as driving behavior, vehicle conditions, road conditions, and environmental factors. The process involves training the EBM model to predict fuel consumption and then generating explanations by calculating the potential fuel reduction if specific features were optimized to reference values (e.g., median values for non-anomalous vehicles). The study also incorporates an anomaly detection step using boxplot analysis to identify vehicles with unusually high fuel consumption. The validity of the model’s explanations is evaluated by comparing the identified influential factors against established domain knowledge and prior literature. The results demonstrate that the EBM model effectively predicts fuel consumption and provides explanations that align with existing scientific understanding. Specifically, 70% of the categories associated with fuel factors identified by the model were consistent with previous literature. The study quantifies the potential impact of optimizing driving behavior, estimating that such improvements can decrease fuel consumption by 12% to 15% in large industrial fleets. The model successfully isolated the individual contributions of various factors, such as aggressive driving maneuvers, tire pressure, and auxiliary system usage, allowing for precise identification of inefficiencies. The significance of this work lies in its application of XAI to a practical industrial problem, bridging the gap between predictive accuracy and interpretability. By providing fleet managers with actionable, quantified insights into which specific behaviors and conditions drive fuel waste, the approach enables targeted interventions to reduce economic costs and carbon emissions. This supports broader environmental goals, such as the Sustainable Development Goals (SDGs), by offering a complementary strategy to vehicle electrification. The study validates the use of EBM in transportation contexts, demonstrating that explainable machine learning can effectively translate complex real-world data into clear, operational strategies for fuel optimization.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• eco driving