Designing and Developing a Model for Detecting Unusual Condition in Urban Street Network

This study introduces a novel model aimed at identifying traffic congestion, accidents, and other irregularities within transportation networks by leveraging a combination of unsupervised machine learning techniques and statistical models applied to speed time series data. The detection of congestion and accidents is a critical aspect of traffic management, yet previous research has seldom focused solely on speed data. This investigation utilizes mobile data sourced from the Neshan application, which operates on the Global Navigation Satellite System (GNSS) and provides 10-min averaged speed metrics. Additionally, data regarding accidents and congestion from the Mashhad traffic control center has been employed to assess and validate the performance of the trained models. The findings of this research yield two distinct methodologies for dissimilarity detection: one that utilizes time series clustering and deviation distribution-based techniques to identify various patterns, and another that employs outlier detection methods, including density-based clustering and predictive confidence bounds, to uncover anomalies. The ensemble model developed demonstrates commendable performance, achieving an accuracy exceeding 80 percent and a robust anomaly detection rate based on the 10-min averaged speed time series. Consequently, the proposed model holds significant potential for the development of real-time anomaly detection applications within urban transportation systems in the future.