Optimizing Passenger Comfort in Rail Transit via Predictive Jerk Estimation Models

Abstract

Increasing train acceleration and deceleration can improve system performance in a railway network. However, passengers are also more likely to lose their balance and collapse. The purpose of this article is to investigate the effects of longitudinal vehicle accelerations on passenger comfort and safety. In addition to the results of previous empirical studies on the maximum acceleration that train passengers can tolerate, the literature review combines two different academic fields to investigate the physiological and kinesiological impacts of acceleration on balance [2]. This Research paper proposes a novel machine learning (ML)-based approach for real-time jerk prediction to enhance train passenger stability. We look into a variety of machine learning methods, including deep learning techniques (particularly, recurrent neural networks, or LSNs). and supervised learning models (Random Forest, Support Vector Machines, and Gradient Boosting), [3] to forecast train dynamics using sensor information such as velocity, acceleration, and position. Our method forecasts short jerk incidents using time-series analysis, allowing for real-time passenger alert systems or proactive changes to train operations. We employ rigorous cross-validation to verify the performance of the mentioned algorithms, evaluating F1-score, accuracy, precision, and recall. The results were outstanding and demonstrate that models based on deep learning, particularly LSTM networks, outperform traditional methods by offering higher prediction accuracy in dynamic scenarios.