Analysis of Feature Correlation for Music Genre Classification

Abstract

Music genre classification has been widely discussed by some researcher. There are various methods used to classify many types of music genres, however only a small part of them considered the importance of feature correlation. This feature correlation is to select features to increase the accuracy of classification process. In this paper, we investigate the big role of features correlation where features are obtained from entropy of root mean square and frequency. Moreover, we use probabilistic neural network (PNN) as the classifier. In this paper, results showed that accuracy using all feature (without considering feature correlation) is obtained 70%, meanwhile using selected features from correlation score, accuracy is conducted 90%. The selected features from this high accuracy are minimum and average RMS entropy of all RMS entropies in each music frame, and minimum and average frequency entropy of all entropies in each music frame.