Data mining applied to automatic music genres classification

Abstract

It aims to classify musical genres automatically by means of Data Mining algorithms, considering descriptors extracted from the audio signal. It identifies the 150 most popular songs of seven musical genres (Rock, Jazz, POP, Classical Music, MPB, Heavy Metal and Samba). By extracting descriptors related to the audio signal of these songs, it applies algorithms Random Forest; Bayes Net; C4.5; KNN and the Bagging and Boosting strategies for the classification. It obtains the best result of 66.67 % of success with the algorithm C4.5 for classification between Samba and MPB. It notes that the classification of musical genres presents itself as an "interesting problem" for studies involving Machine Learning techniques. It stimulates the continuity of similar studies applying algorithms based on Neural Networks and Genetic Algorithms.