EDGE-Intelligent AIoT Framework for Real-Time Traffic Congestion Prediction in Smart Transportation Systems

Abstract

Urban traffic congestion represents a critical challenge for modern cities, causing economic losses exceeding $166 billion annually in the United States alone while contributing significantly to greenhouse gas emissions and reduced quality of life. Traditional centralized traffic management systems struggle with latency, bandwidth limitations, and scalability issues when processing massive data streams from distributed sensors. This research develops and evaluates an edge-intelligent Artificial Intelligence of Things (AIoT) framework that integrates edge computing with deep learning models for real-time traffic congestion prediction in smart transportation systems. The framework deploys lightweight convolutional neural networks and long short-term memory networks at edge nodes positioned near traffic sensors, enabling local data processing and prediction generation with minimal cloud dependency. Through implementation across three urban testbeds encompassing 487 intersections and evaluation using 14 months of traffic data, the framework demonstrates 91% prediction accuracy for congestion events 15-30 minutes in advance, with mean prediction latency of 127 milliseconds—representing 94% latency reduction compared to centralized cloud-based approaches. The system achieves 73% reduction in bandwidth consumption through edge-based data aggregation and filtering, while maintaining resilience during network disruptions with 97% uptime across distributed deployments. Edge intelligence enables dynamic model adaptation responding to local traffic pattern changes, improving prediction accuracy by 23% compared to static centralized models. Energy consumption analysis reveals 68% reduction in total system energy usage through edge processing optimization and intelligent sensor activation. This research contributes a validated architectural framework for edge-intelligent traffic prediction, novel lightweight deep learning models optimized for resource-constrained edge devices, and empirical evidence of performance advantages over traditional approaches, providing practical pathways for cities implementing next-generation intelligent transportation systems.