پیش‌بینی تغییر خط راننده در آزادراه با استفاده از شبکه عصبی LSTM و ماشین بردار پشتیبان (SVM) بر اساس داده‌های میکروسکوپی

Lane changing is one of the most critical driving maneuvers, playing a significant role in both traffic safety and flow efficiency. The aim of this study is to develop and compare machine learning models for short-term prediction of drivers’ lane-change behavior on freeways. For this purpose, high-resolution microscopic data from the Next Generation Simulation (NGSIM) dataset were utilized, and two predictive models—Long Short-Term Memory (LSTM) neural networks and Support Vector Machines (SVM), were designed and evaluated. Predictions were performed for time horizons ranging from 1 to 5 seconds prior to the lane-change maneuver in order to assess the effect of prediction horizon on model performance. Evaluation metrics, including Mean Squared Error (MSE), Recall, and F1-score, were applied to assess model accuracy. The results indicate that the LSTM model significantly outperformed SVM across all evaluation criteria and achieved the highest accuracy with the lowest error in short-term horizons (1–5 seconds). Furthermore, statistical analysis revealed that lane-change maneuvers were more frequent in the right lanes of the freeway. The findings of this study can contribute to the design of advanced driver-assistance systems (ADAS) and the development of accurate predictive algorithms for semi-autonomous and autonomous vehicles.