Intelligent Transportation Systems for Urban Environments: A Hybrid Offline-online Framework for Traffic Prediction and Dynamic Routing

Urban traffic congestion in megacities demands accurate, production-ready prediction systems to enable intelligent navigation and management.This research presents a robust intelligent transportation system that bridges the gap between academic machine learning and real-world deployment through a novel hybrid offline-online framework.The methodology ensures perfect consistency between validation and production via identical 20dimensional spatial-temporal feature engineering and evaluation protocols.Rigorous evaluation on 4.8 million TomTom-derived traffic records demonstrates that Long Short-Term Memory networks significantly outperform Random Forest, achieving 1.85 km/h RMSE (51% improvement) and 99.4% classification accuracy with 82.9% minority-class precision.Visual analysis reveals Long Short-Term Memory superior temporal dynamics versus Random Forest's static predictions around 30 km/h.The production system scales from 41K training segments to 460K operational segments, maintaining 1.73 km/h RMSE, 98.7% routing success, and validated travel timesdemonstrating successful translation of academic research into deployable megacity infrastructure.