Abstract:

The main objective of this study is to explore the achievements of artificial nerve networks from machine learning algorithms, nerve foolish logic control and support vector machines in the country credit rating forecast. Method - The research problem deals with the country’s credit score forecast as a classification problem. Selected variables were used as input for the relevant algorithms and the classification success of these algorithms was studied. The data set used is three-year current data covering the period 2016-2018 of the countries. Results - A large number of trials have been done to select the parameters of the used algorithms and the appropriate parameters have been tried to be identified. According to the research results, both of the three models have been the algorithm that produces the best results for support vector machines, which is a very powerful classification algorithm while the forecast success is high. Discussions - It is possible to say that all the models are acceptable for classification success, they are models that can be used for credit rating determination. For example, in Leshno and Spector (1996) the correct estimated percentage of the YSA model was 72%; in Mohapatra, De, and Ratha (2010) 75%; in Blanco and others (2013) 92.4%. Compared to the studies in literature, the results of the analysis appear to be successful.

Keywords: