A RULE-BASED SCORING METHOD FOR DRIVER BEHAVIOUR EVALUATION USING VEHICULAR PARAMETERS

Abstract

The analysis and evaluation of driver behaviour play a critical role in enhancing road safety, optimising fleet operations, and enabling fair usage-based insurance models. Traditional methods, such as manual observation, often lack objectivity and scalability, limiting their effectiveness in large-scale applications. This study introduces a transparent, rule-based scoring system designed to assess driver behaviour using real- time vehicular data collected via the OBD-II interface. The methodology involved selecting key driving parameters including speed, engine RPM, throttle position, engine load, application of brake, and steering speed based on Sri Lankan government regulations, expert mechanic input, and findings from the literature review. Data comprising approximately 4,000 records were collected from five vehicles driven by different drivers under typical urban and suburban conditions. The dataset was thoroughly pre-processed to remove noise and invalid data points, ensuring consistency and reliability across various car models. Each parameter was assigned thresholds to categorise observed values as good, acceptable, or poor, based on established regulatory standards, expert recommendations, and benchmarks identified through a comprehensive literature review. These scores were weighted according to their relative importance to safety and vehicle health. The total weighted score for each trip was computed by summing the weighted parameter scores and subsequently normalised to a 10-point scale for consistent interpretation. Trips were classified into four behavioural categories: excellent, safe, caution advised, and risky. Application of the scoring system to a dataset of 4,000 trip records demonstrated its capability to effectively differentiate driver behaviour classes. The distribution of results showed that 32.5% of trips were categorised as excellent, 41.0% as safe, 18.0% as caution advised, and 8.5% as risky, reflecting the overall driving behaviour captured across the dataset. To validate the accuracy of these classifications, an additional controlled test was conducted using a separate vehicle driven under deliberately risky conditions, such as high speed and high engine RPM. This test vehicle was correctly classified as Risky by the model, supporting the validity of the categorisation approach. This distribution confirms the system’s ability to distinguish varying levels of driving quality and risk. The system’s transparent and interpretable nature, combined with its independence from large labelled datasets, supports its practical deployment in real-world contexts. The results highlight the system’s potential as a valuable tool for insurers, fleet managers, and road safety authorities.