AI-Based Mobility Prediction Using Recurrent Neural Networks and Attention Mechanisms

Accurate mobility prediction plays a pivotal role in optimizing transportation systems, supporting urban planning, and alleviating traffic congestion. In this work, we present a novel artificial intelligence-based approach for short-term public transport demand prediction, utilizing Recurrent Neural Networks equipped with Long Short-Term Memory units and attention mechanisms. This hybrid framework is designed to effectively model complex temporal dependencies by leveraging the attention mechanism’s capacity to highlight informative patterns across varying time scales. Our method is trained and validated using real-world mobility datasets, consistently outperforming traditional statistical models in terms of prediction accuracy. In our experiments, the model achieved 96.8% accuracy, demonstrating its potential to enhance transport system responsiveness and support personalized mobility services. Furthermore, the integrated attention layer provides valuable interpretability, making the model particularly suitable for real-world deployments in areas such as dynamic fleet management, traffic monitoring, and integration across multimodal transport networks, thereby promoting smarter and more adaptive urban mobility solutions.