Detecting Anomalies in Financial Data Using Machine Learning Algorithms

Bookkeeping data free of fraud and errors are a cornerstone legitimate business operations. The highly complex laborious work financial auditors calls for finding new solutions algorithms to ensure the correctness statements. Both supervised unsupervised machine learning (ML) techniques nowadays being successfully applied detect anomalies in data. In accounting, it is long-established problem misstatements deemed anomalous general ledger (GL) Currently, widely used such as random sampling manual assessment bookkeeping rules become challenging unreliable due increasing volumes unknown fraudulent patterns. To address risk audit inefficiency, we seven ML inclusive deep two isolation forest autoencoders. We trained evaluated our models on real-life GL dataset vectorization resolve journal entry size variability. evaluation results showed that best have high potential detecting predefined anomaly types well efficiently discern higher-risk entries. Based findings, discussed possible practical implications resulting accounting auditing contexts.