Combination of Feature Selection and Optimized Fuzzy Apriori Rules: The Case of Credit Scoring
Seyed Sadatrasoul, Mohammad Gholamian, and Kamran Shahanaghi
Faculty of Industrial Engineering, Iran University of Science and Technology, Iran
Abstract: Credit scoring is an important topic, and banks collect different data from their loan applicants to make appropriate and correct decisions. Rule bases are favourite in credit decision making because of their ability to explicitly distinguish between good and bad applicants. This paper uses four feature selection approaches as features pre-processing combined with fuzzy apriori. These methods are stepwise regression, CART, Correlation matrix and PCA. Particle Swarm is applied to find the best fuzzy apriori rules by searching different support and confidence. Considering Australian and German UCI and an Iranian bank datasets, different feature selections methods are compared in terms of accuracy, number of rules and number of features. The results are compared using T test; it reveals that fuzzy apriori combined with PCA creates a compact rule base and shows better results than the single fuzzy apriori model and other combined feature selection methods. Optimization outperformed FCFS and round-robin algorithms.
Keywords: fuzzy apriori, feature selection, particle swarm, credit scoring.
Received September 7, 2012; accepted December 28, 2013