Feature Vector Design for Music Genre Classification

With the massive growth of digital music availability, it emerges the need of categorising them according to some tags, like the music genre. The development of automatic music genre classification systems have been a research target over the years. This work proposes to investigate the generation of a concise set of problem descriptive feature vectors, a relevant stage when developing most classification systems. It includes a comprehensive study conducted with the GTZAN Dataset, relatively to the quantity and quality of feature vectors necessary for an accurate music genre classification, considering a range of machine learning algorithms, such as the k-nearest neighbours, Multinomial Logistic Regression, Support Vector Machine, Random Forest, and Gradient Boosting. Additionally, the Structured Orthogonal Matching Pursuit, a recent feature selection technique, is evaluated to address this problem, leading to promising results.