Abstract:
Fast population growth and the use of personal vehicles in modern cities challenge the effective operation of traffic in urban centers, as it creates congestion and accident-related problems. Static traffic management systems find it extremely difficult adapting to dynamic conditions, and the challenge finds solution in Smart solutions. In this paper, we introduce a hybrid anomaly detection system to enable real-time traffic monitoring in smart cities that combines the Z-score statistical analysis along with the use of k-Nearest Neighbors (kNN) machine learning to provide a high degree of detection accuracy and efficiency. Z-score is fast at detecting statistical outliers, whereas kNN variant improves the performance of detecting outliers by considering the relationship structure of data points, which minimizes the occurrence of false positives. Its framework is based on the lightweight MQTT protocol to allow real-time interaction between IoT sensors and control systems, which guarantees timely anomaly reports. The hybrid model performed better than standalone Z-score, kNN algorithms in achieving a 97 percent accuracy, 95 percent precision, and 100 percent recall and F1-score on the January 31, 2024, set of events collected in the PeMS District 7 dataset of California. The lightweight, scalable system can be deployed onto resource-limited IoT devices and provides a viable solution to enhance the efficiency of traffic management, safety and responsiveness in urban and other areas.
