Node Based Anomaly Detection with Autoencoders

Abstract

In many systems, individuals are identified and monitored after behavioral issues arise. In this study, anomaly detection is used to identify suspects based on their movement patterns before incidents occur. Unlike other studies, this approach relies on predefined node points and travel times between them instead of visual data. Autoencoder and variational autoencoder models were used to analyze individuals' travel routes, and anomalies were detected based on reconstruction error. The results indicate that the variational autoencoder model produces more precise and accurate results compared to the autoencoder. It demonstrates superior performance in detecting complex anomaly patterns in behavioral movement data. © 2025 Elsevier B.V., All rights reserved.