Rough Set-Based Feature Weighted Kernels for Support Vector Machine

Support vector machine (SVM) is a popular classification paradigm in machine learning and has achieved great success in real applications. The standard SVM neglects the relative significance of each feature with respect to the classification task. Rough set is a valid mathematic tool to handle imprecision, uncertainty and vagueness. Conventionally, the rough set-based feature significance is adopted as heuristic information for feature selection. In this paper, the problem of improving SVM by using feature weighed kernels with rough set-based feature significance is considered. In more detail, the proposed method first estimates the relative significance of each feature by rough set theory, and then utilizes the significance as feature weight to adjust the kernel function in SVM. In this way, the SVM can avoid being dominated by trivial relevant or irrelevant features and lead to an improvement of the generalization performance. The proposed method is demonstrated with some UCI machine learning benchmark examples.