Explainable Hybrid Deep Learning Framework for Real-Time Intelligent Urban Mobility Prediction in Smart Cities
Year: 2026
Type: conference-paper
Abstract: Rapid urbanization and the increasing complexity of transportation infrastructures have intensified challenges related to traffic congestion, mobility inefficiency, fuel consumption, environmental pollution, and urban transportation management in smart cities. Intelligent Transportation Systems (ITS) supported by Artificial Intelligence (AI), Internet of Things (IoT), and deep learning technologies have emerged as promising solutions for improving urban mobility prediction and transportation sustainability. However, many existing AI-based transportation systems still operate as black-box models with limited transparency and interpretability, reducing stakeholder trust and limiting deployment in safety-critical transportation environments. This study proposes an Explainable Hybrid Deep Learning Framework for intelligent urban mobility prediction in smart cities. The proposed framework integrates Attention-based Long Short-Term Memory (Attention-LSTM), XGBoost ensemble learning, and Explainable Artificial Intelligence (XAI) techniques to simultaneously improve prediction accuracy and model transparency. Multi-source transportation datasets consisting of traffic density, vehicle speed, congestion index, GPS trajectories, weather conditions, road occupancy, and temporal mobility indicators were collected from IoT-enabled smart transportation infrastructures. Data preprocessing, feature engineering, temporal sequence generation, and hybrid deep learning optimization were performed prior to model training. SHAP (SHapley Additive Explanations) was utilized to analyze feature importance and interpret prediction behavior. Experimental results demonstrate that the proposed Attention-LSTM-XAI framework achieved the best predictive performance with 97.2% accuracy and the lowest RMSE value compared to Random Forest, XGBoost, GRU, CNN-LSTM, and standard LSTM models. Explainability analysis revealed that traffic density, vehicle speed, and weather conditions were dominant factors affecting congestion prediction. The proposed framework contributes to trustworthy, scalable, and real-time intelligent transportation systems for sustainable smart city mobility management.
Language: English
Subfield: Transportation
Field: Social Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7204765874