Forensic Audit Model Based on Benford’s Law and Anomaly Detection Algorithms in Corporate Expense Reports

Abstract

Corporate expense reports are often vulnerable to fraudulent practices such as fictitious entries, inflated costs, and data manipulation, which can undermine the reliability of financial statements and damage stakeholder trust. Traditional auditing methods, while essential, face limitations in handling large-scale data and in identifying complex patterns of fraud. This research aims to address these challenges by developing a forensic audit model that integrates Benford’s Law with anomaly detection algorithms to improve the accuracy of fraud detection. The study employed anonymized corporate expense report records containing numerical, categorical, and textual attributes. Data preprocessing involved cleaning, normalization, and feature extraction, followed by applying Benford’s Law to identify abnormal digit distributions. Machine learning algorithms, including Isolation Forest, Local Outlier Factor (LOF), and DBSCAN, were then used to detect anomalies in transaction data. The findings indicate that the hybrid model combining these methods achieved superior results with an accuracy of 93.4% and an F1-score of 90.4%, outperforming standalone approaches and significantly reducing false positives and false negatives. This suggests that integrating statistical and machine learning techniques enhances the reliability and efficiency of forensic auditing in detecting suspicious financial activities. The study concludes that such a hybrid framework provides a more comprehensive fraud detection system, offering practical benefits for corporate internal control and regulatory oversight, as well as theoretical contributions to the advancement of data-driven forensic auditing.