On the Use of Feature Selection for Music Genre Classification

In the recent years, utilizing machine learning in musicrelated problems has attracted researchers in both industry and academia. One of the recent targeted challenges is classifying music segments based on their genre, which is done according to the extracted features of their audio tracks. This identification process plays a major rule in the user-tailored recommendation systems employed by the widely used web services like Spotify and YouTube. In this paper, we demonstrate the use of feature selection combined with Support Vector Machine (SVM) classifier to classify the recently shared open-source FMA (Free Music Archive) dataset. We use information-gain feature selection method to select the minimum number of features required for classification without affecting the accuracy of the model. We demonstrate that confining the model to use the top selected features have reduced the model complexity, and significantly reduced the processing time without sacrificing accuracy.