| E. Dritsas, M. Trigka, Ph. Mylonas |
| Comparative Analysis of Unsupervised Anomaly Detection Methods for Big Urban Mobility Data |
| 11th South-East Europe Design Automation, Computer Engineering, Computer Networks and Social Media Conference (SEEDA-CECNSM 2026), Thessaloniki, Greece, September 18-20, 2026 |
|
ABSTRACT
|
| Urban transportation systems generate large volumes of mobility data that can be exploited to identify abnormal travel patterns and operational irregularities. However, anomaly detection in such environments remains challenging due to the absence of labeled observations and the heterogeneous nature of transportation data. This study investigated the use of unsupervised machine learning (ML) techniques for anomaly detection in large-scale New York City (NYC) taxi trip data. A preprocessing and feature engineering pipeline was developed to construct a multivariate representation of trip behavior using mobility, temporal, and economic attributes. Three anomaly detection algorithms, namely, Isolation Forest (IF), Local Outlier Factor (LOF), and One-Class Support Vector Machine (OC-SVM), were evaluated. The experimental results showed that although all models identified the same anomaly rate under a common contamination setting, substantial differences were observed in the detected anomaly sets. The strongest agreement was found between IF and OC-SVM, whereas LOF identified a considerably different subset of anomalous observations. The analysis of the detected anomalies revealed recurring patterns associated with unusual fare-distance relationships, atypical travel durations, and abnormal trip costs. The findings demonstrate that different anomaly detection paradigms capture complementary aspects of abnormal mobility behavior and highlight the effectiveness of unsupervised learning for anomaly discovery in large-scale urban transportation data.
|
| 18 September, 2026 |
| E. Dritsas, M. Trigka, Ph. Mylonas, "Comparative Analysis of Unsupervised Anomaly Detection Methods for Big Urban Mobility Data", 11th South-East Europe Design Automation, Computer Engineering, Computer Networks and Social Media Conference (SEEDA-CECNSM 2026), Thessaloniki, Greece, September 18-20, 2026 |
[
BibTex] [
Print] [
 Back] |
