امیرحسین حاجی محمدیزدی

The present comparative study investigates the performance of three data-mining methods to predict the type of auditor's opinion regarding the financial statements of stock exchange companies. The database consisted of the information of ۴۲۰ company-years collected between ۲۰۱۸ and ۲۰۲۱, which included ۳۰ financial and non-financial variables. In this study, decision tree (DT), support vector machine (SVM), and K-nearest neighbor (K-NN) techniques were used to develop auditor's opinion prediction models. Additionally, the logistic regression (LR) method was used to compare the performance of the mentioned techniques with >The results revealed that the three investigated techniques perform better in predicting the auditor's opinion than the logistic regression. Although these three data mining techniques showed high accuracy in predicting the type of opinion, the K-nearest neighbor (K-NN) method showed the highest performance in terms of prediction accuracy with ۹۱.۷۳% and the minimum errors of the first type (alpha risk) and the second type (beta risk), both by ۴.۱۳%. According to the values of the errors and the coefficient of sensitivity, and clarity that their combination can be observed in the ROC curve, the efficiency of these techniques was also tested, and the mentioned method showed the best results. Additionally, DT techniques (with a difference of about ۳۱%) and SVM (with a difference of about ۱۲%) perform better in predicting acceptable comments than conditional comments. However, the KNN method (with a difference of about ۲%) has a suitable and desirable capability in predicting both types of comments and works very similarly.

Keywords: Data mining, Auditor's opinion, Decision tree (DT), K-nearest neighbor (K-NN), Support vector machine (SVM), Logistic regression (LR).