Abstract: Often, most of products to be inspected are qualified, and only a small fraction is unqualified. In other words, the number of qualified and unqualified products are highly unbalanced. In the identification of criticaltoquality characteristics (CTQ), significant performance deviation is observed when traditional method is applied. The performance of identifying CTQ for the unqualified products is significantly inferior to that for the qualified products. In order to solve problem, improved information gain (IG) algorithm is proposed to process such highdimension imbalance data. By this method, it reduces the influence of imbalance data on the performance such that the identification of CTQ is significantly improved. Numerical simulation for an example verifies the effectiveness of the proposed method.