K-Means Analysis for Music Segmentation in Playlist Recommendation

Abstract

Nowadays, music streaming platforms are the main choice for music lovers to enjoy songs digitally. To improve user experience, a recommendation system is needed that is able to suggest personalized music based on individual preferences. This research examines the application of the K-Means Clustering algorithm in audio feature-based music segmentation to generate musically and emotionally relevant playlists. Spotify was chosen as the sample in the study because it is the largest and most widely used music streaming platform in the world. The dataset is obtained from Kaggle and contains audio attributes such as tempo, energy, valence, and danceability. After preprocessing and normalizing the data, the K-Means algorithm was used to group songs into clusters based on the similarity of their audio features. The clustering produced four distinct song groups, each characterized by specific audio traits such as tempo, energy, and valence. These clusters were then mapped into playlist themes, such as Energetic & Uplifting and Mellow & Relaxing, based on their dominant features. In a limited user test involving ten listeners, the playlists were rated positively in terms of musical flow and emotional coherence. Notably, Cluster 2 received the highest satisfaction scores. These results indicate that clustering songs based on audio features is not only technically viable but also enhances the emotional alignment of music recommendations.