<![CDATA[This study identifies patterns in popular Spotify songs based on measurable audio characteristics and evaluates the relationship between the resulting clusters and stream-count tendencies. The Most Streamed Spotify Songs 2023 dataset was used, consisting initially of 953 records and 24 attributes, with 948 songs retained after data cleaning. Eight audio features—bpm, danceability, valence, energy, acousticness, instrumentalness, liveness, and speechiness—were used for clustering, while streams was analyzed only after clustering. Preprocessing included outlier examination and comparison of RobustScaler, StandardScaler, and MinMaxScaler. The number of clusters was evaluated using the Elbow Method and Silhouette Score, with k = 2 selected as the final configuration. The K-Means model achieved a Silhouette Score of 0.9218, a Davies-Bouldin Index of 0.3284, and a Calinski-Harabasz Index of 2857.8019. Cluster 0 contained 930 songs and showed higher danceability, valence, energy, and speechiness with very low instrumentalness, whereas Cluster 1 contained 18 songs and was primarily distinguished by much higher instrumentalness. Sensitivity analysis showed that the imbalanced structure was stable across centroid initializations but strongly dependent on instrumentalness and the selected scaling method. Comparison with Agglomerative Clustering and Gaussian Mixture Model showed that K-Means offered competitive clustering quality with the lowest observed runtime. Although Cluster 0 had a higher mean stream count, the Mann-Whitney U test found no statistically significant difference between clusters (p = 0.3495). The final solution is therefore interpreted as an algorithmically stable but feature-dependent separation, while stream differences are treated as descriptive rather than causal.]]>
