Özet:

Financial statement fraud is when companies change and publish the data in their financial statements in line with their interests. Detecting financial statement fraud that causes serious damage to organizations, stakeholders, and the economic structure is an important problem. There are various control mechanisms for this. However, there is a need for innovative control methods against new fraud methods that may be developed over time. Data mining is a promising field for detecting financial statement fraud. In the classification analysis of data mining, patterns are obtained from existing data with classification methods, and these are used in the classification of unseen units. In this study, research was carried out on the identification of companies with the risk of financial statement fraud through the classification methods of data mining. The data were obtained from the financial statements published by companies traded in Borsa Istanbul between 2014 and 2018. Seven classification methods were used first, and the three most successful ones were selected. In the next stage, hyperparameter optimization was carried out to improve the performance values. Accuracy rates of 91.73 percent with K-Nearest Neighbor, 90.51 percent with Random Forest, and 90.37 percent with XGBoost were obtained from the classification methods, and the best prediction rate was obtained with K-Nearest Neighbor. In the last part, the best performance values were also obtained with K-Nearest Neighbor in the comparisons using random sub-sampling method.

Anahtar Kelimeler: