This study compares the ability of four data-mining techniques in the prediction of audit opinions on companies' financial statements. The research data consists of 37,325 firm-year observations for companies listed on the New York Stock Exchange (NYSE), the American Stock Exchange (AMEX), and the NASDAQ from 2001 to 2017. The dataset consists of U.S. companies' various financial and non-financial variables. This study uses Decision Trees (DT), Support Vector Machines (SVM), K-Nearest Neighbors (K-NN), and Rough Sets (RS) to develop the prediction models. While all models developed by these four techniques predict the audit opinions with relatively high accuracy, the SVM models developed by the RBF kernel demonstrate the highest performance in terms of overall prediction accuracy rates and Type I and Type II errors. The results indicate that all models developed using different algorithms demonstrate their highest performance in predicting going-concern modifications, ranging from 84.2 to 100 percent.

JEL Classifications: M42.

You do not currently have access to this content.