Research on Intelligent Audit System Based on Optimized BP Neural Network

The higher level of complexity in organizational and financial auditing requires cognitive systems to classify audit results accurately and efficiently. The research suggests a hybrid model of optimized a Backpropagation (BP) Neural Network for improving prediction accuracy based on the Audit dataset. The data set, consisting of categorical and numeric features pertaining to organizational and transactional properties, is preprocessed initially using normalization and categorical encoding to achieve compatibility with the neural design. PSO is utilized as a metaheuristic optimizer to optimize the initial weights and biases of the BP network so that the local minima and slow convergence problems normally faced by gradient descent-based training are avoided. The optimized BP model then classifies to detect patterns and anomalies in the audit records, enhancing the accuracy of risk assessment. Experimental results show that the optimized BP model has a classification accuracy of 97.6 %, far higher than the traditional BP neural network. The method is of. considerable potential for the automation of audit processes, saving human efforts, and enhancing decision-making in financial and compliance monitoring systems.