Detection of Anomalous Vehicular Traffic and Sensor Failures Using Data Clustering Techniques

2025-04-01

AI Summary

This study addresses the joint detection of traffic anomalies (e.g., sudden congestion) and sensor faults in highway sensor time-series flow data. We propose a distance-metric-based, clustering-driven anomaly detection framework that systematically compares and integrates symbolic aggregate approximation (SAX) and dynamic time warping (DTW) representations. To our knowledge, this is the first work to rigorously evaluate the robustness of multiple clustering paradigms—including hierarchical clustering, k-means, and fuzzy c-means—in traffic anomaly diagnosis. Experiments on real-world highway sensor data demonstrate substantial improvements in detection accuracy, effectively distinguishing genuine traffic anomalies from sensor failures, with false positive rates reduced by up to 23.6%. Our core contribution is a novel, interpretable, multi-granularity-adaptive clustering–anomaly joint diagnostic paradigm, offering a principled pathway toward enhancing the reliability of intelligent transportation sensing systems.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Planning, Routing, and Scheduling: Activity and Plan Recognition

Application Category

• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Security and Privacy: Large-scale security measurements

Abstract

The increasing availability of traffic data from sensor networks has created new opportunities for understanding vehicular dynamics and identifying anomalies. In this study, we employ clustering techniques to analyse traffic flow data with the dual objective of uncovering meaningful traffic patterns and detecting anomalies, including sensor failures and irregular congestion events. We explore multiple clustering approaches, i.e partitioning and hierarchical methods, combined with various time-series representations and similarity measures. Our methodology is applied to real-world data from highway sensors, enabling us to assess the impact of different clustering frameworks on traffic pattern recognition. We also introduce a clustering-driven anomaly detection methodology that identifies deviations from expected traffic behaviour based on distance-based anomaly scores. Results indicate that hierarchical clustering with symbolic representations provides robust segmentation of traffic patterns, while partitioning methods such as k-means and fuzzy c-means yield meaningful results when paired with Dynamic Time Warping. The proposed anomaly detection strategy successfully identifies sensor malfunctions and abnormal traffic conditions with minimal false positives, demonstrating its practical utility for real-time monitoring. Real-world vehicular traffic data are provided by Autostrade Alto Adriatico S.p.A.

Problem

• Detect anomalous vehicular traffic using clustering techniques
• Identify sensor failures in traffic data networks
• Analyze traffic patterns with hierarchical and partitioning methods

Innovation

• Clustering techniques for traffic anomaly detection
• Hierarchical and partitioning methods combined
• Distance-based anomaly scores methodology