Credit quality assessments using manifold based semi-supervised discriminant analysis and support vector machines

Abstract

Due to the large scale of financial data in credit quality forecasting, dimensionality reduction is a key step to enhance classifier performance. By using manifold based semisupervised discriminant analysis (SSDA) and support vector machines, this study develops a novel prediction system for credit quality assessment, where SSDA makes efficient use of labeled and unlabeled (testing) data points to gain a perfect low dimensional approximation of data manifold and simultaneously maintain the discriminating power. More specifically, the labeled data points are used to maximize the separability between different classes, and the testing data points are used to estimate the intrinsic geometric structure of the data space. Empirical results indicate that SSDA outperforms other dimensionality reduction methods with a significant performance improvement, and our hybrid classifier substantially outperforms other conventional classifiers.