Analyzing Drivers' Lateral Vehicle Control Under Various Driving Conditions Using Machine Learning

Abstract: Driver behavior plays a crucial role in road safety, significantly influencing the overall safety of roadways globally. This study examines how terrain, traffic conditions, and driver demographics such as age, gender, and driving experience influence drivers’ lateral vehicle control errors and overall driving performance. Utilizing an in-house developed driving simulator, the experimental design simulates driving on a two-lane undivided rural road with hilly and plain terrains, both with and without traffic, to assess lateral vehicle control errors. A total of 70 participants (50 males, 20 females), aged 18 - 48, were recruited to drive in this controlled virtual environment. The participants were divided into novice and experienced driver categories based on their driving history. Each participant completed multiple driving tasks in the simulated environment, allowing us to collect data on variables such as speed, gear usage, steering angle, lateral and longitudinal acceleration, and collision counts.

We applied a combination of statistical methods such as Analysis of Variance (ANOVA) and post hoc tests and machine learning technique Random Forest modeling to assess the influence of these variables on lateral control errors. The results revealed that terrain and traffic conditions significantly affected driver performance. Hilly terrains and high-traffic conditions were associated with a notable increase in lateral control errors, particularly among novice drivers. Gender also played a significant role, with female drivers exhibiting higher collision counts in challenging conditions.

Notably, age emerged as the most important factor in predicting lateral control errors, with younger and older drivers showing more frequent errors than middle-aged drivers. Unlike real-world driving, where drivers rely on both visual and mechanical feedback (lateral accelerations), the simulator lacked physical feedback, which may have skewed the performance of more experienced drivers. This limitation highlights the challenges of replicating real-world driving behavior in a controlled, virtual environment.

Despite these limitations, the Random Forest model provided valuable insights into the most significant predictors of lateral control errors. Age, terrain type, and vehicle speed were found to be the most critical factors influencing driving performance, accounting for over 70% of the variance in collision counts. These findings underscore the need for targeted interventions, such as tailored training programs for novice and older drivers, especially in challenging driving environments. These findings highlight the importance of developing targeted interventions, including customized training programs for novice and older drivers, especially in challenging driving conditions