Transfer Learning Approach to Debt Portfolio Appraisal

Abstract

Machine learning and data mining algorithms usually assume that the training and future data have the same distribution and come from the same feature space. However, in majority of real-world problems, this is not true. In case of Debt portfolio appraisal we have sufficient training data only in another domain of interest, namely in other portfolios. Therefore, only knowledge transfer from these portfolios in inference for new one is possible. In the paper we propose transfer learning and learning based on similarity methods, basing on similarity between training and testing datasets. The proposed approach is examined in real domain debt portfolio valuation.