Evolvable hybrid ensembles for musical genre classification.

The possibility of classifying musical songs according to their musical genre is becoming more challenging because of the millions of songs included in online databases. Therefore, reliable and efficient methods need to be developed that will automatically solve this task. In this article, the mentioned task is accomplished using sets of classifiers. The authors' contribution to the development of automatic musical genre recognition is using hybrid ensembles formed from deep neural networks and classical classifiers and the optimization process executed on the voting process of individual classifiers. Finally, differential evolution algorithms have been used to improve the classification quality further. The proposed evolutionary algorithm shows improvement in comparison with other optimization methods.