Financial information manipulation prediction: Comparison of statistical methods and machine learning algorithms

In this study, performances of probit and logit analyses, which are traditional statistical methods, and support vector machine, decision tree, random forest, k-nearest neighbor and artificial neural network methods, which are machine learning algorithms, are compared in predicting financial information manipulation. The classification accuracies of all methods are above %60, and the method with the highest classification accuracy is random forest and the lowest is support vector machine. Machine learning algorithms have been found to be more successful than traditional methods. The biggest disadvantage of these algorithms is that probability values cannot be obtained. Probit and logit analyses are needed to obtain probability values. In the analyses, in addition to the Beneish model which is frequently used in the literature to predict financial information manipulation, revaluation, operating expenses, other income and expenses from main activities indices, which are thought to be effective in predicting financial information manipulation in Turkey, have been used. The indices that are effective in predicting financial information manipulation have been determined. Gross profit, sales growth, leverage, marketing, selling, distribution and general administrative expenses, operating expenses, other income from main activities indices and the ratio of accruals to assets have been found to be statistically significant in predicting financial information manipulation.