Credit Risk Analysis Using Sparse Non-negative Matrix Factorizations

Abstract

Credit risk analysis is to determine if a customer is likely to default on the financial obligation. In this paper, we will introduce sparse non-negative matrix factorization method to discovery the lower dimensional space for reducing the data dimensionality, which will contribute to good performance and fast computation in the credit risk classification performed by support vector machine. We test the sparse NMF in a real-world credit risk prediction task, and the empirical results demonstrate the advantage of sparse NMF by comparing with other state of art methods.