Monitoring Fuel-Efficient Driving Patterns to Augment ADAS to regulate the fuel dynamically using Machine Learning

Abstract

In the realm of enhancing Advanced Driver-Assistance Systems (ADAS) for optimal vehicle performance and sustainability, this paper introduces an innovative methodology that capitalizes on Machine Learning (ML) to scrutinize and learn from fuel-efficient driving patterns. Utilizing real-time driving cycle datasets collected through On-Board Diagnostics (OBD) tools, meticulously recorded the driving styles of 100 drivers on a predetermined route over duration of 35 minutes. The rich dataset includes an array of vehicle parameters and driving behavior, providing a comprehensive foundation for ML-based analysis. Proposed approach involves the application of state-of-the-art ML algorithms, with a specific focus on Machine learning and ensemble methods, to accurately model and predict fuel-efficient driving patterns. The algorithms XGB-regression, Random Forest regression and Gradient Boosting regressions were rigorously trained, validated, and tested on the collected data, ensuring robustness and reliability in their predictive capabilities. The scores and the Mean absolute Errors of these algorithms were estimated, most efficient algorithms are RF-regression and GB-regressions with a score of 99.93% and 99.93% with MAE 0.01219% and 0.01210%respectively. The integration of these predictive models into the ADAS framework mayshow promising results, significantly improving the system’s performance in real-time decision-making and driver assistance. The system demonstrates an enhanced capability to adapt to varying driving styles, offering personalized recommendations for fuel-efficient driving, and contributing to the reduction of fuel consumption and emissions.