Music genre classification using radial basis function networks and particle swarm optimization

This work presents the development of an intelligent system able to classify different music genres with increased accuracy. The proposed approach is based on radial basis function (RBF) networks, trained with the non-symmetric fuzzy means particle swarm optimization-based (PSO-NSFM) algorithm. PSO-NSFM, which has been shown to produce highly accurate regression models, is in this case suitably tailored to accommodate for classification problems. The classifier's performance is evaluated using the Matthews correlation coefficient (MCC), which can better reflect the success rate per individual class, by summarizing the entire confusion matrix. The resulting classification scheme is applied to the well-known GTZAN dataset, where the objective is to classify 10 different musical genres, based on half-minute music audio excerpts. A comparison with different classifiers shows that the proposed approach offers improved classification accuracy.