Financial Data Anomaly Detection Method Based on Residual Explanation for Government Office Operational Expenditure

Abstract

Ensuring fiscal responsibility and transparency in government offices requires careful control of operating expenditures. However, due to the intricacy and regularity of financial transactions, this work can be difficult, making standard monitoring techniques inefficient and time-consuming. This paper introduces an anomaly detection method for financial data in government office operational expenditures, termed "Residual Explanation". This approach uses sophisticated machine learning techniques to discover anomalous transaction patterns that can point to fraud by analyzing residuals, or the disparities between observed and predicted transaction values. Our method makes use of a Random Forest Regressor model, which is especially well-suited to managing the high dimensionality and non-linear correlations seen in financial datasets. This study focuses on anomaly detection within operational expenses of account 525112 at one of institutions in Indonesia. The results indicate that the optimal model for detecting anomalies operates with a thresholds value of 99,6%. Future improvements to this model could allow its integration of our approach into existing financial systems could enable real-time anomaly detection, which is paramount for preventing fraud and enhancing the financial governance of government expenditures.