Supervised learning and unsupervised learning on music data with different genres
Chunang Liu, Zehan Chao · 2021
As the digital music datasets evolve fast in the past decades, researchers start to focus on the machine learning tasks on music datasets with music content analysis. Both supervised learning and unsupervised learning have become increasingly popular as searching and recommending are the major tasks for digital music datasets. In this work, we are interested in exploring automatic classification of music with different genres. In this paper, we visualize and embed Spotify music in 3 dimension space with principal component analysis techniques; classifying different music genres via MFCC related algorithms and various machine learning algorithms. We use the method of PCA to determine the relationship between each feature of those songs in the data including 154932 songs and utilize different visualization techniques to obtain insight into the dataset. Thus, we could get the similarity of the songs by analyzing the two clusters of points on the graph. Additionally, we use numerous supervised learning techniques to perform classification on those songs in ten genres and evaluating their performance. Experimental results show the feasibility of automatic management of music databases and the potential to improve.