Financial statement analysis represents a fundamental component of audit procedures, requiring extensive examination of numerical data, trends, and relationships across multiple reporting periods. Traditional audit approaches rely heavily on manual analytical procedures and rule-based testing, leading to time-intensive processes and potential inconsistencies in analysis depth and coverage. The increasing complexity of financial reporting and growing volumes of financial data have intensified these challenges. This study proposes a Deep Learning (DL) framework designed to automate and enhance financial statement analysis in audit contexts. The framework integrates Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) to analyze financial statement patterns, detect anomalies, and identify potential misstatements. Advanced deep learning algorithms process multi-period financial data to recognize complex relationships and unusual variations that may indicate audit risks. Experimental validation using financial statements from 500 public companies demonstrates that the proposed framework achieves 89.7% accuracy in anomaly detection and reduces analytical procedure time by 73%. The system successfully identifies potential misstatements and unusual fluctuations while maintaining high precision rates. Implementation results show significant improvements in audit analytical efficiency, consistency, and risk identification capabilities.
Edwards et al. (Fri,) studied this question.