Integrated intelligent systems for corporate credit analysis

Abstract

Some of prior researches using artificial intelligent techniques for bankruptcy prediction and the prediction of bond rating did not show outstanding performance partly because of the tremendous noise of financial data. Those researches mostly paid little attention to the importance of preprocessing. The other reason of inconsistence of performance of prior researches was due to the multi-collinearity characteristics of input variables of the prediction model. Accordingly selecting adequate input variable according to specific data set is inevitable for designing prediction model. If relevant data and variables are appropriately selected for modeling the system, then the noisy of information are eliminated and the prediction model may show better prediction accuracy. In part II, this dissertation present a hybrid data mining model for the prediction of corporate bond rating. This model uses a new case-indexing method of case-based reasoning (CBR), which utilizes the cluster information of financial data in order to improve prediction accuracy. This method uses not only case-specific knowledge of past problems like conventional CBR, but also uses additional knowledge derived from the clusters of cases. The cluster-indexing method assumes that there are some distinct subgroups (clusters) in each rated group. Competitive artificial neural networks like self-organizing map (SOM) and learning vector quantization (LVQ) are used to generate the centroid values of clusters because these techniques produce better adaptive clusters than statistical clustering algorithms. The experiments using corporate bond rating cases show that the cluster-indexing CBR is superior to conventional CBR and inductive learning-indexing CBR - a rival case indexing method. In addition, the majority models of cluster-indexing CBR augment the predication accuracy of the cluster-indexing CBR after excluding the minor cases out of each cluster. In part III, this dissertation presents an applicat...