An AI-Driven IoT Solution for Public Transport Quality Monitoring and Driving Style Estimation

Elevating passenger satisfaction in public transit is crucial for incentivizing sustainable mobility and mitigating urban greenhouse gas emissions. To address this challenge, this study introduces an innovative Transport Quality Monitoring System (TQMS) framework that synergizes Internet of Things (IoT) and Artificial Intelligence (AI) to collect and evaluate both empirical and user-perceived quality metrics systematically. The proposed TQMS features a tri-centric architecture spanning hardware, edge, and cloud levels, embedding human-centric feedback directly into the digital replication of the service. Operationally, vehicular telemetry captured via dedicated on-board sensors provides objective insights into driving dynamics, while qualitative passenger feedback is gathered via smartphones. This heterogeneous data is subsequently synthesized at the Edge Layer by the proposed IoT solution running intelligent analytics for real-time quality assessment and anomaly detection. The system's performance was evaluated through empirical field tests conducted on operational urban bus lines, involving 60 participants, 1,200 ride segments, and 20 lines. For predictive modeling, we present a one-vs-rest neural ensemble optimized with a novel temporal coherence loss function designed to minimize jitter in consecutive estimation intervals. Empirical results, obtained under subject-independent, route-independent, and combined subject-and-route evaluation protocols with fold-wise feature selection to prevent data leakage, demonstrate that this tailored loss function on the collected dataset consistently improves upon standard class-weighted Binary Cross Entropy (BCE), increasing macro-F1 by 0.06–0.07 and enhancing the weighted-average F1-score for overall transport service and driving style by up to 8.2% across five distinct quality dimensions, with the largest gains observed on the least represented quality classes.