Audit Risk Identification based on Mutual Information with Backpropagation Neural Network
Di Su · 2025
In recent years, identifying audit risk had become important for maintaining financial transparency and detecting fraud. Instead of focusing solely on risk valuation, there had been an increased demand for advanced audit systems. The existing methods suffer from irrelevant feature inclusion, slow convergence in audit risk classification. Therefore, this research proposes Mutual Information based Backpropagation Neural Network (MI-BPNN) for identifying audits risks. Initially, audit data dataset is taken which incorporated audit data information of various firms. The collected audit is preprocessed by utilizing data cleaning which removes errors and normalization for ensuring uniform. Then this audit data features are selected by employed MI method for selecting the most informative indicators. The selected relevant features are evaluated for detecting the fraudulent firms by utilizing BPNN method. The proposed MI-BPNN achieved better results which comprises accuracy of (98.58%), specificity of (97.95%), sensitivity of (96.69%), and recall of (97.20%) when compared with existing Long Short-Term Memory (LSTM).