A Literature Review of Statistical and AI-Based Metro Passenger Demand Forecasting Models

1.Purpose
This study contributes to the NEXUS project, an EU-funded initiative focused on enhancing metro operations through AI, data science, and advanced forecasting. Recognizing the challenges posed by fluctuating passenger demand, disruptions, and evolving passenger behaviours, this research explores improved demand forecasting methodologies to support adaptable and sustainable metro systems.

2.Research Approach
The research conducts a comprehensive review and analysis of both traditional statistical methods and emerging AI-based models for passenger demand forecasting in metro operations. Quantitative approaches including ARIMA, SARIMA, probabilistic regression, and deep learning models such as neural networks, generative adversarial networks (GANs), and graph-based attention networks are examined. These methods are evaluated for their applicability across different forecasting horizons and operational contexts, with particular attention to their performance under data-limited or expansion scenarios.

3.Findings and Originality
While traditional statistical models remain effective for short- and medium-term forecasting where historical data is available, they face limitations in dynamic environments such as new station planning and network expansions. In contrast, AI-based models especially those leveraging deep learning, hybrid approaches, and alternative data sources demonstrate superior adaptability and predictive performance in complex, non-linear, or data-sparse contexts. This work advances the NEXUS objective by identifying innovative techniques that can enhance forecasting accuracy and operational decision-making in modern metro systems.

4.Research Impact
The findings directly support ongoing NEXUS pilot studies in Genoa (Italy) and Sofia (Bulgaria), providing methodological foundations for integrating AI into real-time demand forecasting, timetable optimization, and operational planning. The research informs the development of adaptive control frameworks capable of responding to fluctuating demand with improved resource allocation, service quality, and passenger satisfaction.

5.Practical Impact
The research will make metro operations more flexible and responsive to demand, reducing overcrowding and underutilised services. Improved vehicle utilisation and increased capacity will lead to cost savings and energy efficiency. The study aims to optimise passenger flow, reduce dwell times, and improve service quality, promoting inclusivity, accessibility, and reducing reliance on private vehicles.