Comparison of data mining methods for audit opinions prediction: An application in Borsa Istanbul

This study compares the performance of twelve distinct data mining methods for estimating audit opinions on companies' financial statements. The dataset comprises 2,093 company-year observations from 161 companies listed on Borsa Istanbul between 2010 and 2022. Independent audit opinion classification is conducted utilizing a set of 28 financial and non-financial independent variables. The methods employed in the study include Bayesian Belief Network, Naive Bayes, Logistic Regression, Artificial Neural Networks, Radial Basis Function, Support Vector Machines, K-Nearest Neighbor, AdaBoost.M1 Algorithm, Decision Trees (J48), Random Forests, Decision Stump, and Classification and Regression Tree for constructing prediction models. Based on the findings, the Random Forests model demonstrated the highest prediction accuracy performance in forecasting audit opinions, achieving a rate of 96.68%. The statistical comparison of the models encompassed prediction accuracy, classification matrix, detailed accuracy results, Type I error rate, Type II error rate, and performance criteria. To this end, pioneers the development of models for predicting accurate audit opinions based on 2,093 company-year observations utilizing both financial and non-financial variables. It is anticipated to contribute to the audit literature by employing data mining classification methods for predicting audit opinion types. Furthermore, the framework design utilized in this study offers a decision support tool for internal and independent auditors, accountants, shareholders, company managers, tax authorities, public institutions, individual and institutional investors, stock exchanges, law firms, financial analysts, credit rating agencies, and the banking system.