Optimized Predictive Modeling of Lane-Specific Vehicle Time Gaps for Traffic Flow and Road Management

Lane-specific characteristics, traffic flow, vehicle properties, and atmospheric conditions profoundly impact successive vehicle time intervals. A lane position plays an important role due to the differences in flow properties, such as speed, density, and overtaking maneuvers, which can vary significantly across different lanes. Larger or heavier vehicles, which are assigned to specific lanes, tend to require although time more space from each other due to their dimensions and braking requirements. Driver behavior, as determined by reaction times and compliance with rules, introduces variability, especially in challenging weather or road conditions, such as rain, snow, or icy roads. In addition to geographical location and distance of travel, temporal factors also influence perception and choice, including the day of the week and variations in light intensities. Effective modeling approaches and effective traffic control require a comprehensive analysis of these factors. In the current work, a dataset of independent traffic roads with all the above-mentioned variables was constructed, and time-gap prediction between two consecutive cars driving along every road lane was performed using Support Vector Regression (SVR) and Random Subspace (RSS). In addition, new optimization algorithms, referred to as the Partial Reinforcement Optimizer (PRO) and the Walrus Optimizer (WO), were utilized to enhance predictive capability, resulting in strong hybrid models. In this evaluation approach, we also employed Analysis of Variance (ANOVA’s) sensitivity analysis technique to identify the most contributing feature in our data, providing us with the importance of each variable in the prediction process. The SVPR hybrid model performed the best for lane 1, with the highest R² of 0.997, the lowest Root Mean Square Error (RMSE) of 2.36E+07, and the lowest Ratio of RMSE to Standard deviation (RSR) of 0.059 during testing, thereby reflecting its superior predictive accuracy and minimal error. For lane 2, the hybrid SVWO model emerged as the most effective, achieving the highest R² of 0.989, the lowest RMSE of 3.36×10^7, and the smallest RSR of 0.107, demonstrating its robust capability in capturing lane-specific traffic dynamics. These findings highlight the potential of hybrid optimization techniques to enhance predictive performance and minimize errors in practical traffic management systems.