Analyzing Credit Card Defaulters: A Comparative Study Using Kohonen Maps, Neural Networks, and Decision Trees

The text discusses the credit card debt crisis faced by credit card issuers in Taiwan, which has led to a decrease in consumer financial confidence and poses problems for both banks and cardholders. There is a study conducted using data on customers who took credit cards in order to identify the most important criteria for non-payers. The study aims to predict loan repayment ability using a neural network and selecting the most significant factors. The literature review highlights the use of neural networks in determining non-payers and the potential for creating systems to warn credit card owners about risk behavior. The methodology involves data processing using the Kohonen map, a type of neural network that uses unsupervised learning for visualization and clustering. Decision Trees and Neural Networks are also used in the analysis. Analysis using Kohonen maps did not provide the necessary information due to heterogeneous data in clusters. Analysis via a neural network resulted in a high error percentage of 17.6^% and lacked accuracy in predicting average values. Analysis through the decision tree yielded a lower error percentage of 12.5% and identified repayment status in September and August as the most significant factors. An experiment was conducted by dividing factors into logical groups, resulting in a difference of 7% in error percentages for each group. The conclusion suggests that the decision tree method is the most suitable for this data model, with repayment status in September as the key factor in predicting defaulters.