Audit Risk Identification based on Back Propagation Neural Network with Lasso Regularization

In recent scenarios, the audit risk identification is a critical component of external auditing for the auditors to focus on high-risk areas and ensure accurate financial reporting. However, the existing Risk Assessment Framework (RAF) faced challenges in identifying audit risks due to imbalanced audit risks. Hence, this research proposes a Synthetic Minority Oversampling Technique with Back Propagation Neural Network (SMOTE-BPNN) to balance the audit risks. The proposed SMOTE-BPNN balances the audit risks by oversampling high-risk audits as well as identify them by learning patterns and relationships among the risks. Initially, the input data is collected from Audit dataset and then fed into processing. Here, the data is preprocessed with data cleaning to eliminate irrelevant attributes and min-max normalization to scale feature values into a specific range. After that, Mutual Information (MI) technique is employed to measure the dependency among variables for the selection of optimal features. Then, SMOTE balances the data by oversampling them and finally BPNN is introduced to identify the audit risks efficiently. From the results, the proposed SMOTE-BPNN achieved better results in terms of Accuracy (96.57%), specificity (94.05%), and sensitivity (93.62%) when compared to existing Bi-directional Long Short-Term Memory with Attention Mechanism (BiLSTM-AM) respectively.