Predicting personal credit ratings using ubiquitous data mining

Ubiquitous data mining (UDM) is a methodology for creating new knowledge by building an integrated financial database in a ubiquitous computing environment, extracting useful rules by using diverse rule-extraction-based data mining techniques, and combining these rules. In this study, we built six credit rating forecasting models using traditional statistical methods (i.e., logistic regression and Bayesian networks), multilayer perceptron (i.e., MLP), classification tree algorithms (i.e., C5.0), neural network rule extraction algorithms (i.e., NeuroRule), and UDM in order to predict personal credit ratings. To verify the feasibility and effectiveness of UDM, credit ratings and credit loan data provided by A Financial Group in Korea were used in this study. Empirical results indicated that UDM outperforms other single traditional classifiers such as logistic regression, neural networks, frequency matrix, C5.0, and NeuroRule. UDM always outperforms other single classifiers in credit rating forecasting; it can predict future personal credit ratings more accurately than any other single classifier. Keywords—Ubiquitous data mining (UDM), Ubiquitous computing environment, Credit rating forecasting models, Rule extraction algorithms, Integrated financial database.