2026· International Journal of Computer Science and Engineering Innovations· 0 citations
TL;DR
A generic framework is presented which exploits the zero-shot, few-shot and multi-modal capabilities of foundation models to forecast traffic flow, predict traffic incidents and improve public transit schedules.
Abstract
Intelligent Transportation Systems (ITS) are undergoing a paradigm shift driven by the unprecedented volume of multi-modal data and the necessity for highly accurate, real-time predictive analytics. Traditional machine learning and deep learning architectures, while effective for localized tasks, often struggle with spatial-temporal distribution shifts, cross-city generalization, and the integration of heterogeneous data sources such as traffic cameras, GPS trajectories, weather reports, and textual incident logs. This paper investigates the application of Foundation Models (FMs) large-scale architectures pre-trained on massive datasets and fine-tuned for downstream tasks as the core engine for next-generation ITS predictive analytics. The article presents a generic framework which exploits the zero-shot, few-shot and multi-modal capabilities of foundation models to forecast traffic flow, predict traffic incidents and improve public transit schedules. This study uses large-scale comparisons and empirical simulations to evaluate the structural benefits, integration techniques, and performance trade-offs of FMs against common deep learning frameworks. The results show that foundation model-based methods achieve better generalization on unseen urban networks and high robustness towards the noisy or missing sensor data. Lastly, we identify some of the principal two-sided challenges including computational overhead, latency constraints in edge deployment and privacy protection, also discuss a few strategies for addressing these issues and direction towards new research and deployable practices within smart city infrastructures.
A key contribution of this study lies in its comparative synthesis of ML and DL models, revealing that hybrid and graph-based DL architectures consistently outperform traditional ML methods when handling large-scale, heterogeneous traffic datasets.
Thabo Matue, A. A. Akinyelu, Mase Mokotsolane· International Journal of Dat...· 0 citations
Accurate short-term traffic prediction is a critical component of intelligent transportation systems (ITS), yet it remains challenging due to nonlinear temporal dynamics, evolving spatial dependencies, and uncertainty in real-time urban traffic data. This paper proposes a novel uncertainty-aware deep ensemble spatiotemporal forecasting framework integrating Dynamic Graph Convolutional Networks (DGCN), Temporal Transformers, and CNN–LSTM hybrid models. A confidence-guided ensemble fusion strategy dynamically weights individual predictions using Bayesian uncertainty estimation. Experiments conducted on real-time Bhopal city traffic data demonstrate significant improvements over state-of-the-art baselines, achieving up to 90% performance gains during peak and abnormal traffic conditions.
An integrated prediction-and-visualisation pipeline that transforms complex data distributions into actionable visual analytics, such as interpretable station-to-station demand heatmaps via interactive GIS Folium layers is implemented, providing an operationally robust framework to support smart-city transportation management and build more sustainable urban transit systems.
Berna Çalışkan· Journal of Data Analytics an...· 0 citations
TETRA is proposed, a hybrid spatio-temporal traffic forecasting model that integrates Graph Convolutional Networks (GCNs) with Extended Long Short-Term Memory (xLSTM) to capture complex multi-timescale temporal patterns, including congestion propagation and delayed recovery dynamics, which are not well represented by conventional recurrent models.
Norman Bereczki, Vilmos Simon· International Journal of Int...· 0 citations
Precise short-term travel demand forecasting and personalized trip advice are two of the most challenging problems in intelligent transportation systems in part due to the fact that traveler behavior is influenced concurrently by spatial configuration, time dynamics, socio-demographic environment, and cost sensitivity. The majority of the current models cover these dimensions separately. The current paper suggests an end-to-end, monolithic framework that pairs the learning of multimodal representations with a spatio-temporal graph neural network (ST-GNN), a Transformer-based forecasting module, a causal inference layer, and a Q-learning recommender. Heterogeneous travel history, such as demographics, trip purpose, mode of transport, cost of accommodation, and temporal attitudes are co-embedded and propagated through graph convolutions and multi-head self-attention to generate multi-horizon destination-level demand prediction. An actionable do-calculus layer measures the impact of traveler characteristics on decisions and the policy implications guide a reinforcement learning agent that modulates transport and accommodation suggestions to traveler groups. The ST-GNN + Transformer forecaster achieves a 25% lower RMSE than an ARIMA baseline on a publicly available dataset of approximately 8,700 traveler records and, in comparison to the LSTM and DCRNN alternatives, converges within just 300 training episodes. The framework provides a replicable base to adaptable, understandable model of travel behavior in smart-city and tourism analytics contexts.
Santosh Kumar Sharma, Satish Chander, Piyush Gupta· International journal of com...· 0 citations
As transportation networks grow increasingly complex and data-rich, the need for intelligent, adaptive routing mechanisms has become essential for efficient and resilient mobility operations. This study presents a deep learning–enabled framework for real-time dynamic route optimization in logistics systems, addressing fundamental limitations of traditional static routing and heuristic-based decision approaches. The proposed architecture integrates long short-term memory (LSTM) networks with spatio- temporal graph convolutional networks (ST-GCN) to model nonlinear temporal evolution and spatial dependencies in traffic flows, GPS trajectories, meteorological conditions, and road network structures. By capturing these complex patterns, the predictive module generates highly accurate short-term forecasts of congestion levels and delivery delays, which are subsequently incorporated into an adaptive routing engine that continuously updates vehicle paths in response to evolving network conditions. Comprehensive preprocessing of multimodal traffic and environmental datasets, advanced feature engineering, and supervised training of the LSTM and ST-GCN models are employed. Model performance is assessed via mean absolute error (MAE), root mean square error (RMSE), and ROC–AUC. Experimental results show substantial gains over baseline predictors and conventional routing: a 45.6% reduction in MAE, a 39.5% reduction in RMSE, and an ROC–AUC of 0.91 for delay prediction, while enabling an estimated 12.3% reduction in carbon emissions. These improvements translate into measurable reductions in travel time and fuel consumption, underscoring the system’s potential to enhance operational resilience, environmental sustainability, and decision efficiency.
Ahmed Abdel-Wahab Rakha, Mohammed S. A. Elsersy· Informatica· 0 citations