In the face of growing global economic uncertainty, financial auditing has become essential for ensuring regulatory compliance and preventing systemic financial risks. Traditional manual auditing methods are increasingly challenged by large data volumes, complex business structures, and evolving fraud tactics. To address these issues, this study explores an AI-driven financial audit framework and high-risk identification system for enterprises, leveraging machine learning to enhance audit efficiency and accuracy. Using a dataset from the Big Four accounting firms (EY, PwC, Deloitte, KPMG) spanning 2020 to 2025, the research analyzes trends in risk assessment, compliance violations, and fraud detection. The dataset includes critical indicators such as audit project counts, high-risk cases, detected frauds, and compliance breaches, while also reflecting the role of AI in audit automation, employee workload, and client satisfaction. To build a robust risk prediction model, three machine learning algorithms—Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbors (KNN)—are compared. SVM leverages hyperplane optimization for complex classifications, RF integrates multiple decision trees to handle nonlinear, high-dimensional data with strong resistance to overfitting, and KNN uses distance-based classification for adaptable performance. Through hierarchical K-fold cross-validation and evaluation via F1-score, accuracy, and recall, Random Forest demonstrates the highest performance with an F1-score of 0.9012, especially excelling in detecting fraud and compliance anomalies. Feature importance analysis highlights audit frequency, historical violations, employee workload, and client ratings as key risk predictors. The study suggests that enterprise audit systems adopt Random Forest as a core model, extend data features through feature engineering, and implement real-time monitoring. This research offers valuable insights into using machine learning for intelligent financial audits and risk management in modern enterprise environments.
Yuan et al. (Wed,) studied this question.