국내기사
군 재정분야의 금전사고 예방을 위한 비지도학습 기반 이상거래 탐지모델 연구 = A study on an unsupervised learning-based anomaly detection model for preventing financial misconduct in military budget management
This study presents an unsupervised learning-based anomaly detection model aimed at preventing financial misconduct in military budget management. While current rule-based monitoring systems can identify simple irregularities, they are limited in detecting complex or evolving fraud patterns. Using over 220,000 real transaction records from the military financial system, this study trained models on normal data and generated synthetic anomalies representing repeated small transfers to acquaintance accounts. Two unsupervised algorithms, Isolation Forest (IF) and Local Outlier Factor (LOF), were compared. The IF model achieved superior results (Precision = 0.544, Recall = 0.999, F1-score = 0.701), demonstrating high accuracy and robustness in imbalanced data environments. The findings confirm that AI-based anomaly detection can enhance internal control and reduce the risk of financial incidents in military operations, suggesting practical applicability even within closed defense networks.