IoT-Driven Destination Prediction in Smart Urban Mobility: A Comparative Study of Markov Chains and Hidden Markov Models

Abstract

The increasing availability of IoT-enabled mobility data and intelligent transportation systems in Smart Cities demands efficient and interpretable models for destination prediction. This study presents a comparative analysis between Markov Chains and Hidden Markov Models applied to urban mobility trajectories, evaluated through mean precision values. To ensure methodological rigor, the Smart Sampling with Data Filtering (SSDF) method was developed, integrating trajectory segmentation, spatial tessellation, frequency aggregation, and 10-fold cross-validation. Using data from 23 vehicles in the Vehicle Energy Dataset (VED) and a filtering threshold based on trajectory recurrence, the results show that the HMM achieved 61% precision versus 59% for Markov Chains (p = 0.0248). Incorporating day-of-week contextual information led to statistically significant precision improvements in 78.3% of cases for precision (95.7% for recall, 87.0% for F1-score). The remaining 21.7% indicate that model selection should balance model complexity and precision-efficiency trade-off. The proposed SSDF method establishes a replicable foundation for evaluating probabilistic models in IoT-based mobility systems, contributing to scalable, explainable, and sustainable Smart City transportation analytics.