Audit Opinion Prediction with Data Mining Methods: Evidence From Borsa Istanbul (BIST)

This study compares the performance of 12 different data mining methods in predicting audit opinions for companies’ financial statements. The study dataset consists of 2,093 firm-year observations from 161 companies listed on Borsa Istanbul for 2010-2022. The independent audit opinion types were classified using a set of 28 independent financial and non-financial variables. The following prediction models were used in the study: Bayesian Networks, Naive Bayes, Logistic Regression, Artificial Neural Networks, Radial Basis Function, Support Vector Machines, K-Nearest Neighbor, AdaBoost.M1 Algorithm, Decision Trees (J48), Random Forest, Decision Stump, and Classification and Regression Tree (CART). According to the results of the analysis, the Random Forest model demonstrated the best performance with a prediction accuracy rate of 96.68% for predicting audit opinions. The statistical results of the models were compared based on prediction accuracy, confusion matrix, detailed accuracy results, Type I error rate, Type II error rate, and performance measures. This study pioneers developing models capable of accurately predicting audit opinions using 2,093 firm-year observations based on financial and non-financial variables. This contributes likely to the audit literature by predicting audit opinion types through data mining classification methods. The framework design used in the study is anticipated to serve as a decision support tool for internal and external auditors, accountants, shareholders, company executives, tax authorities, other public institutions, individual and institutional investors, stock exchanges, law firms, financial analysts, credit rating agencies, and the banking system when making decisions.