A practical strategy for acquiring rules based on rough sets and principal component analysis

In order to obtain the attribute reducts and the concise rules with stronger generalization capabilities, we propose a practical strategy for acquiring rules based on rough set (RS) and principal component analysis (PCA), called here PSAR-RSPCA. In the PSARRSPCA, the collective correlation coefficient (CCC), as a quantitative index based on the essence of PCA, is used to measure the contribution of every condition attribute to "cause" (i.e. the state space constructed by the entire condition attributes), and RS is developed to keep "causality" (i.e. the dependencies between condition attributes and decision attributes) unchanged in a decision table. Meanwhile, PSAR-RSPCA absorbs the evolution ideas of gene algorithm and stimulated annealing algorithm to search for the attribute reduct with larger CCC. Compared with other algorithm, the test results show PSAR-RSPCA has an obvious reduction in the error rates of prediction (approximately 34.5%) by the well-known classification benchmark.