Artificial intelligence approaches and an application in determining credit risk

Artificial intelligence methods have already survived the infancy and even the walking periods, and it has now become the focus of all circles who want to take advantage of the benefits of technology and thus do their jobs in a smoother, more efficient and creative formation. The solutions created by these methods and approaches that are rapidly developing day by day to ancient problems such as credit risk valuation are among the most curious and researched topics. In this thesis, among the artificial intelligence methods, Artificial Neural Networks (ANN) and Support Vector Machines (SVM), and from traditional statistical methods, Logistic Regression (LR) were used. The methods are applied to three different real life data from a bank. The first of these is the "Individual Data Set" consisting of individual data and 407 observations. The second is the "Corporate Data Set", which contains corporate customer data and consists of 12921 observations. The third is the "Balanced Corporate Data Set", which is also derived from the secondary data set, consisting of 1590 defective and 1590 flawless customer data and 3180 observations in total. In all three data sets, SVM was the method that gave the highest accuracy. When the transition from the corporate data set to the balanced corporate data set, the amount of data decreased by approximately 75%. It was observed that this decrease in data amount did not affect SVM at all (1% increase), ANN (22% decrease) and LR (30% decrease) in terms of accuracy rate. Therefore, in the case of limited data, it has been seen in this thesis that SVM is a more successful method compared to other methods. Keywords: Determination of credit risk, artificial neural networks, support vector machines, control of model complexity