Finansal Tablo Hileleri’nin Makine Öğrenmesi Yöntemleri ve Lojistik Regresyon Kullanılarak Tahmin Edilmesi: Borsa İstanbul Örneği

Abstract

This study aims to create an effective model to predict one year before whether 88 firms, continuously traded at Borsa Istanbul between 2000-2019, commit fraud in their financial statements. For this purpose, financial statement fraud was predicted by using Artificial Neural Networks (ANN), Classification and Regression Trees (CART) and Support Vector Machine (SVM) and Logistic Regression (LR) methods among machine learning methods. As a result, the overall prediction accuracy of ANN (96.15%), CART (96.15%), SVM (80.77%) and LR (80.77%) test samples were obtained. ANN and CART classified correctly in test samples all 13 firms that fraudulent in their financial statements. This shows that all methods used in this study, can be used in studies to predict financial statement fraud. (English) [ABSTRACT FROM AUTHOR]