ANALYSIS OF METHODS FOR FORECASTING URBAN PUBLIC TRANSPORT MOVEMENT

Abstract

The article presents the results of a study the problem of forecasting urban public transport movement under conditions of increasing mobility demand, growing traffic intensity and the need to improve intelligent transport systems. Accurate prediction of vehicle arrival time, travel time between stops, service headways, and vehicle occupancy is essential for route optimization, timetable planning, and improving passenger service quality. The research is motivated by the fact that transport systems operate in a highly dynamic environment influenced by congestion, weather conditions, infrastructure disruptions, and fluctuations in passenger demand.
Analyzed the main methodological approaches used for forecasting urban public transport movement. These include statistical time-series methods (particularly regression analysis, moving average techniques, and ARIMA/SARIMA models), filtering approaches based on the Kalman filter, machine learning and deep learning methods, neural network architectures, hybrid and spatio-temporal models. Particular attention is paid to the strengths and limitations of each approach, their data requirements, computational complexity, and suitability for real-time applications in urban transport systems.
The results show that statistical models remain useful for baseline forecasting due to their simplicity, transparency, and low computational cost, but their ability to capture nonlinear and rapidly changing transport conditions is limited. Kalman filter are effective for short-term real-time estimation when streaming GPS and sensor data are available. Machine learning, deep learning, and graph-based neural network models demonstrate higher predictive accuracy because they can process large-scale heterogeneous data and represent complex temporal and spatial dependencies. Hybrid models that combine statistical and artificial intelligence methods achieve the best overall performance in highly dynamic urban environments.