A Study on Self-Organizing Maps and K-Means Clustering on a Music Genre Dataset

Understanding the relationship of the different music genres have been a research pursued by many. A common approach for it is through clustering. Data within a cluster are more closely related to each other as opposed to data on other clusters. Through the use of SOM, clustering could be easily represented since SOM force clustering on a 2 dimensional plane, however the drawback to this is that SOM take a lot of time to calculate and learn. Another method to cluster is through the use of K-Means, which is computationally much faster than SOM, however more than 3 clusters would make the K-Means a little bit harder to visualize in terms of plotting them due the fact that the K also represents the number of dimensions. Though some studies have already provided means to handle visualization of multi-cluster or multi-dimensional planes, a table is usually still used to determine their relationships. This study focuses on understanding the difference of SOM and K-Means as well as to try to implement in determining the relationships of music genres.