Assessment of Probability Defaults Using K-Means Based Multinomial Logistic Regression

Abstract

Classification analysis is a key and easy tool in machine learning and prediction. Because of the large amount of data and the need to convert this data into useful information and knowledge, machine learning has gotten a lot of attention in the information industry and also in society because of the large amount of data and the issues that come with it. In this paper, a K-Means based Multinomial Logistic Regression (MLR) prediction algorithm is used for evaluating the performance of Probability Defaults (PD), and suggestions are made to improve financial status. The necessary information about the members of PD has been collected from the UCI machine learning repository. The parameters are chosen for the study using the feature selection method. The research goal is to find default risk probabilities and they are assessed by accuracy, RMSE (Root Mean Squared Error), error rate, and time. K-means based Multinomial Logistic Regression (MLR) significantly outperforms other classifier models. Assessment of PD will have an impact on the financial industry.