The Nearest Subspace Method for Assessing Credit Risk

Abstract

In this study, we use a classification strategy called the closest subspace approach to assess credit risk. Identifying "good" and "bad" creditors via credit risk assessment is a common categorization challenge. There has been a lot of talk lately about using machine learning techniques like support vector machine (SVM) to assess credit risk. Yet there is plentyNo tried-and-true pattern recognition or AI-based classification techniques for use in assessing creditworthiness exist. This work proposes using the closest subspace classification technique, a robust approach to facial recognition, in the context of credit scoring. When evaluating creditworthiness, the nearest subspace credit evaluation method uses the subspaces spanned by creditors in the same class to extend the training set, with the Euclidean distance between a test creditor and the subspace serving as the similarity measure for classification.