A heuristic-based rough set features optimization algorithm for compressed audio

We investigate feature selection methods, which have been applied to automatic kinds of compressed audio classification systems. It is based on attribute dependency for feature optimization and modified SVM (Support Vector Machine) for classifier. In this paper, we present a new method for feature selection based on priori knowledge by removing both irrelevant and redundant features, and it still retains sufficient information for classification purpose. Experiments on compressed audio category classification indicated that when using this proposed method to select the optimal feature subset and combing with the modified SVM classifier, we could get better efficiency up to 90%, even 10% higher to the total feature sets.