FUZZY MULTI-CRITERIA RANDOM SEED AND CUTOFF POINT APPROACH FOR CREDIT RISK ASSESSMENT

Data mining classification techniques have been studied extensively for credit risk assessment. Existing techniques by default uses 0.5 as the cutoff irrespective of datasets and classifiers to predict the binary outcomes, thus limiting their classification performance on imbalanced group sizes of datasets. This paper addresses two key problems with the existing techniques and talks about the advantages of using Multiple Criteria Decision Making (MCDM) technique on multiple evaluation criteria. The first key problem is applying default cutoff irrespective of datasets and classifiers. The second one is utilizing single criteria for evaluating classification performance and predicting cutoff point. This research work identifies the best cutoff point with respect to datasets and classifiers and integrates MCDM under fuzzy environment in all data mining stages of evaluation to take better decisions on multiple criteria, selection of initial random seed in the clustering phase for better cluster quality and Best Seed Clustering combined Classification (BSCC hybrid algorithm) with selected features to improve classification performance. The integration of these techniques gives a better hand to improve cluster quality and classification performance score with respect to datasets and classifiers because the cutoff point varies from dataset to dataset and classifiers to classifiers. Experimental outcomes from applied credit dataset of UCI machine learning repository found to be competitive and the proposed BSCC hybrid algorithm increases the performance score on obtained cutoff point over non-hybrid approach with default cutoff.