Feature Selection in Credibility Study For Finance Sector

Abstract With the advances in the Information Technology field, banks can evaluate the credit requests of the customers via effective analytical methods and risk analysis. The software products, named Credit Scoring Systems, consist of collecting customer data based on pre-determined credit factors, processing the data with various statistical or machine learning methods, and conducting a credit risk analysis to make the final credit decision. The effects of the quantitative values of the properties formed in the data set vary according to the results. Determination a subset set of columns with a high impact on the outcome and meaningful and removal of irrelevant columns without effect according to the values in the features of a dataset is called the property selection. It is generally used for accuracy and scaling. In this study, it was studied to determine the areas that most impacted the credit result on the sample dataset to meet the need of the structure that can apply the credit to the consumers who apply for the loan and manage the assessment in the consumers. Information Gain and Gain Ratio algorithms were used to determine the most useful features. As a result of the study conducted on the data set, the valuable values of function were committed by using the Gain Ratio, and Information Gain algorithms, and the characteristics were listed according to their magnitude.