Interpretable machine learning for audit planning: Improving misstatement and compliance risk detection in financial services

This research investigates the usefulness of artificial intelligence-based risk scoring in the planning phase of the audit conducted in controlled industries in comparison with time-tested risk assessment frameworks. The main goals involve the assessment of AI models (gradient boosting with SHAP) predicting financial misstatements, look-up of the non-compliance and operational inefficiencies, analysis of the key features of the audit cycle, and displaying the risk measures with the help of Power BI. The design is a retrospective cohort study based on historical audit data and the metrics applied to evaluate the performance of the models are precision, recall, lift, and false-positive cost. The major results suggest that AI models can substantially improve audit accuracy and efficiency, especially when it comes to risk identification, and appear to be weak regarding recall and false-positive expenses. Heatmaps and other visual tools were discovered to be helpful in making decisions. The study will help to enhance the practice of audit since it will offer actionable alternatives regarding how financial institutions can utilize AI in enhancing the risk management system, including offering recommendations to how organizations can make the most out of their audit planning efforts.