| Abstract: |
The rapid escalation of urbanization has intensified vehicular density across metropolitan road networks, creating unprecedented challenges in traffic congestion, environmental degradation, road safety, and commuter productivity. Conventional fixed-timer signal systems are fundamentally inadequate for managing dynamic, non-linear traffic patterns characteristic of modern smart cities. This empirical study presents the design, implementation, and rigorous evaluation of an Intelligent Traffic Management System (ITMS) that synergistically integrates Internet of Things (IoT) sensor networks with advanced Machine Learning (ML) algorithms to achieve real-time, adaptive traffic control. A heterogeneous IoT infrastructure comprising ultrasonic vehicle detectors, infrared cameras, RFID tag readers, GPS probe modules, and environmental sensors was instrumented across twelve signalized intersections in a mid-sized urban corridor over a continuous six-month observation window. The resulting dataset, encompassing over 2.8 million timestamped traffic observations across vehicular volume, queue length, average speed, intersection occupancy, and ambient conditions, was processed through edge computing nodes and transmitted via LoRaWAN and 5G communication protocols to a centralized cloud analytics platform. Six supervised and deep learning models Logistic Regression, Support Vector Machine, Decision Tree, Random Forest, Long Short-Term Memory (LSTM), and a novel Hybrid CNN-LSTM architecture were trained, validated, and benchmarked for congestion prediction and optimal signal phase duration estimation. The Hybrid CNN-LSTM model achieved the highest predictive accuracy of 94.2%, with 93.5% precision and 92.8% recall, outperforming the best baseline (Random Forest at 88.6%) by 5.6 percentage points. Field deployment of the proposed system resulted in measurable improvements: average vehicle waiting time reduced by 38.2%, intersection throughput increased by 27.4%, fuel consumption decreased by 22.1%, and CO₂ emissions dropped by 18.7% compared to the legacy fixed-timer system. Statistical significance was confirmed via paired t-tests (p < 0.001). These findings demonstrate the transformative potential of ML-driven IoT architectures in building sustainable, responsive, and intelligent urban transportation ecosystems and provide a replicable empirical framework for smart city planners and transportation engineers. |
