Presenting a comprehensive model for predicting the type of audit opinion from machine learning algorithms: Evidence from Tehran Stock Exchange

The main aim of the represented research is to provide a comprehensive model for predicting the type of audit opinion based on a number of machine learning algorithms in some companies in the Tehran Stock Exchange. In order to achieve this goal, 1,606 company-years (146 companies for 11 years) observations collected from the annual financial reports of companies admitted to the Tehran Stock Exchange from 2010 to 2020 have been tested. In this study, six machine learning algorithms (decision tree and regression, random forest, neural network, nearest neighbor, logit regression, support vector machine) and also two methods of selecting the final variables of the research (two samples mean comparing test, forward step-by-step selection method) has been used for the model creation. The results show that the overall accuracy of decision tree and regression, random forest, neural network, nearest neighbor, logit regression, and support vector machine procedures respectively are 78.7%, 77.7%, 76.9%, 74.6%, 78.3%, and 76.7%. Regarding the obtained outcomes, the decision tree and regression algorithm outperform in forecasting the type of audit opinion compared to other studied methods. Meanwhile, in general, the result of variable selection techniques illustrates that the step-by-step method is far more effective. Hence, in the studied companies in the Tehran Stock Exchange, the step-by-step method and the decision tree and regression algorithm provide the most efficient model for the prediction of the audit opinion type.