익명의 사용자 이동 패턴 학습과 지리적 이상감지 연구

Abstract

Personal safety and crime prevention have become pressing societal concerns. While wearable devices such as smartwatches offer features including Global Positioning System (GPS) tracking and emergency alerts, their ability to proactively recognize deviations from a user's usual path is limited. This study proposes a Long Short-Term Memory (LSTM) based trajectory learning algorithm that leverages anonymized user data without additional identifiers. It enables detection of changes from a user DB of usual trajectories and thus allows recognition of anomalies in real-time. Experimental results demonstrate that the model achieves relatively consistent performance in predicting distance errors for paths, although the time prediction performance may vary depending on path characteristics. In anomaly detection analyses, normal paths maintained stable values without exceeding the set threshold, while anomalous paths exhibited increasing error values over time, eventually exceeding the threshold.