<![CDATA[This study aims to cluster songs based on listener activities using the Self Organizing Map (SOM) algorithm with Spotify audio features. The dataset used in this study was obtained from Kaggle with a total of 20,718 songs. The variables used include Danceability, Energy, Valence, and Acousticness. The research stages consist of data pre-processing, searching for the best SOM parameters using Grid Search, SOM clustering, clustering evaluation using Silhouette Score and Davies-Bouldin Index (DBI), activity labeling, and implementation of a Streamlit-based web dashboard. The results show that the best SOM parameters were obtained using a 2x3 grid, 3000 iterations, sigma 1.0, and learning rate 0.5 with a Silhouette Score of 0.2348 and a DBI value of 1.2894. The Silhouette Score indicates that the separation between clusters is moderate but not perfect, which is reasonable because music data often have continuous and overlapping audio characteristics. Meanwhile, the DBI value indicates that the similarity between clusters is still relatively high, although the clusters can still be interpreted based on their dominant audio characteristics. The clustering process produced six activity clusters, namely Workout, Dancing, Gaming, Sleeping, Studying, and Hanging Out. The Gaming cluster became the largest cluster with 31.35% of the data, while the Workout cluster became the smallest with 9.53%. The web dashboard implementation successfully visualized clustering results, music activity distributions, cluster characteristics, and music recommendations based on listener activities. The study concludes that the SOM algorithm is capable of clustering songs based on Spotify audio feature similarities at a moderate level and can be implemented in activity-based music recommendation systems.]]>
