Exploring and Mining the Boundaries of Genre Complexity

Online music streaming has become extremely ubiquitous amongst society. In the last year, we have seen over 600 million individuals subscribe to a music streaming platform. With user experience playing a large role in purchasing and trusting a product, music genre classification becomes ever more important. Music genres provide streaming services a way to recommend new music to its users which can serve to improve their overall user experience. With the amount of music added to these platforms every day automatic music genre classification continues to be a focal point of research. Therefore, our objective is to contribute to the literature research of automatic music genre classification. In this study, we seek to evaluate the performance of a kNN music genre classifier with an increasing number of target genre classes. Our methodology involves training and testing five kNN classifiers each with an increasing number of target genres. From these tests we evaluate the performance of each model using metrics such as accuracy, precision, and recall. The results of this study suggest that the performance of a kNN classifier decreases with an increase in the number of target classes. The highest accuracy model is the classifier with two target classes at an accuracy of 94% while the least accurate model contains six target classes with an accuracy of 55%.