Explainable Artificial Intelligence for Journal Entry Anomaly Detection and Continuous Financial Auditing

Abstract

The financial audit is proving to be holding its own in the sense of necessity to have intelligent mechanisms which not only enable to identify aberrations in the journal entries but also provide a clear and understandable evidence on the basis of audit decision. In this paper, an Explainable Artificial Intelligence (XAI) architecture of Journal Entry Anomaly Detection and Continuous Financial auditing is presented, an architecture merging data preprocessing, feature engineering, anomaly detection, explainability, risk assessment and continuous audit monitoring into one architecture. The framework analyzes properties of transactions such as time of posting, relationship between the account and the user, amount of transaction, user, document and historical online accounting trends to identify suspicious journal entries. Models used to detect anomalies are created to produce their own risk score and explainability mechanisms are used to interpret the factors behind each of the flagged transactions. The proposed framework also has control based on accounting rules and continuous monitoring which helps in the facilitation of early identification of the potentially suspicious financial activities. The given features, risk types and description summaries of high-priority transactions are displayed in an interactive audit layer, allowing the auditor to examine the anomaly scores. The framework evaluation is interested in the efficiency of detection, consistency of the explanations, reduction of false-positive and efficiency of audit response. The framework will promote transparency to the audit, accountability, and ongoing financial control and reduce the use of largely-manual transaction screening, with this methodology which combines automated anomaly detection with easily-understandable evidence