Autonomous Vehicle Mode Shift’s Effect on Traffic Efficiency: A Comparison of Dedicated Lane and Mixed Traffic Approaches
Abstract
Developing countries, including Indonesia, are characterized by Heterogeneous Disordered Traffic (HDT), making traditional technology adoption models for autonomous vehicles (AVs) unsuitable. To bridge this gap, this study surveyed 204 urban early adopters using a rigorous psycho-statistical framework. Utilizing t-tests, ANOVA, and bootstrapped ordinal regression, we validated three behavioral pillars: safety-first prioritization, educated critical adoption, and a direct mode shift to shared AVs. Subsequent market segmentation via a CHAID decision tree yielded empirical AV market penetration rates (MPR) of 20%, 40%, and 70%. MPRs were evaluated in a calibrated PTV VISSIM microsimulation of the A.P. Pettarani Arterial Road in Makassar, Indonesia. A tree-based ensemble framework (XGBoost, CatBoost, and Random Forest) with SHAP diagnostics and a Double Machine Learning causal protocol assessed infrastructure trade-offs. With an R2 of 0.99 and MAPE of 9.9%, XGBoost strongly predicted intersection delays and road section speeds. Causal analysis shows that putting dedicated lanes in place too soon, when the penetration rate is low (≤20% MPR), makes traffic worse. At an intermediate 40% MPR, dedicated lanes reduce intersection bottlenecks. Mixed non-dedicated traffic wins at 70% widespread saturation. The AV fleet acts as macroscopic pacemakers, smoothing traffic shockwaves and reducing intra-lane oversaturation to maximize global network efficiency without spatial segregation.