چکیده:
In most of the research conducted, predictor variables have been selected without criteria and only based on past studies. The process of selecting variables can be considered as a pre-processing stage to remove irrelevant variables and add and select optimal variables before creating the model. In this regard, the present research investigates the usefulness of a correlation-based variable selection method for predicting the type of auditor's opinion of companies listed on the Tehran Stock Exchange. The classifiers of this research are artificial neural networks and logistic regression. In general, the empirical findings related to the examination of 1214 observations (year - company) in the time range of 1386 to 1393 showed the usefulness of using the variables selected by the correlation-based variable selection method in the performance of predicting the type of auditor's opinion. In other words, if the variables selected by this method are used compared to using all the initial variables, the average accuracy will increase and the type I and type II errors will decrease. In addition, the findings of the research indicate the appropriate and better performance of neural networks compared to logistic regression.
خلاصه ماشینی:
In this regard, the present research examines the usefulness of a correlation-based variable selection method for predicting the type of auditor's opinion of companies accepted in the Tehran Stock Exchange.
Predicting the type of auditor's opinion, logistic regression, correlation-based variable selection method, neural networks 1.
On the other hand, findings from past research indicate that the audit report, its type and content, carries significant informational load and economic consequences and can affect stock price or return (Chen & Zhaou, 2000; Shabahang & Khatami 1377; Hassas Yeganeh & Yaqubi Mansh, 1382), bankruptcy prediction (Shourvarzi, Bakhtiari, Zenddel & Esmaeilzadeh Bagh Siahhi, 1390), manager compensation (DeAngelo, 1981) and credit grantor decision-making regarding granting credit (Firs, 1980).
Thus, the main issue of this research is to investigate whether selecting optimal predictor variables from the initial variables has a positive and significant impact on the prediction performance of the auditor's opinion of companies listed on the Tehran Stock Exchange.
Considering the importance of selecting predictor variables, this research examines the advantage of the correlation-based variable selection method on the prediction performance of the auditor's opinion of companies listed on the Tehran Stock Exchange.
Many studies have been conducted to predict the type of auditor's opinion using other artificial intelligence methods such as support vector machines (Dampous, Gaganis & Passiouras, 2005); probabilistic neural networks (Gaganis, Passiouras & Dampous, 2007); nearest neighbor (Gaganis, Passiouras, Spatis & Zopounidies, 2007); decision trees (Karkas et al.
(
, 
, 
)
