Rough set (RS) theory gives an approach to knowledge acquisition. Most RS-based reduction algorithms are to find minimal attribute reducts with heuristic knowledge derived only from the dependencies between condition attributes and decision attributes. The rules based on minimal attribute reducts are too simple to represent expert knowledge well. To acquire rules with stronger generalization capabilities, we advocate a knowledge acquisition approach based on RS and Principal Component Analysis (KA-RSPCA). KA-RSPCA uses a collective correlation coefficient as heuristic knowledge to assist attribute reduction and attribute value reduction. The coefficient is a PCA-based quantitative index to measure every condition attribute's contributions to the state space constructed by the whole of the condition attributes in a decision table.

For two test data sets in comparison with other algorithms, KA-RSPCA algorithm shows a reduction in error rate with only a slight increase in the number of rules used.