International Journal of Science, Engineering and TechnologyAuthors: Dr. C. Venish Raja, Mohana M
Abstract: Public transport in fast-growing cities is under constant pressure, with buses and metro coaches becoming overcrowded at peak hours while remaining underused at other times. Passengers often have no reliable way to know how crowded a vehicle will be before it arrives, while fixed timetables and static journey planners cannot respond effectively to changing demand. Artificial intelligence provides an alternative by treating crowding as a time-series forecasting problem and using predicted occupancy as an input to route recommendation. Research reviewed in this paper reports that recurrent deep-learning models such as Long Short-Term Memory (LSTM) and bidirectional LSTM (BiLSTM), together with tree-based ensembles such as Random Forest and XGBoost, can outperform traditional statistical and schedule-based approaches for occupancy forecasting. Integrated systems that connect prediction with fleet decisions and passenger recommendations are also reported to provide greater operational benefits than isolated prediction models. To confirm that the problem is relevant from the passenger perspective, a primary survey of 123 public transport users was conducted. The survey indicates that overcrowding is regularly experienced by a substantial share of respondents, that many respondents have changed their route because of crowding, and that willingness to trust an AI-based crowding system is divided. These findings support the need for transparent and explainable recommendations rather than an unexplained black-box score. This paper presents the main approaches to AI-based crowding prediction, reports the primary survey findings, explains important feature families and model types, describes a route recommendation method that considers predicted crowding together with travel time, waiting time and transfers, discusses real-time deployment and robustness, and proposes an evaluation plan. The proposed direction combines a data layer, BiLSTM and gradient-boosting prediction models, an ensemble mechanism, and a multi-objective recommendation layer. The overall objective is to support less crowded and efficient journeys while helping transport operators balance capacity. The paper also identifies challenges involving unusual events, new routes, sensor failures, privacy, fairness, passenger behaviour and technology acceptance, and recommends future work on adaptive, lightweight and privacy-aware systems.
