<![CDATA[K-Means is one of the most widely used clustering algorithms because of its simplicity and computational efficiency. However, its performance often decreases when handling non-linear data due to the assumption that all attributes contribute equally to the distance calculation process. This study proposes a Variance-Weighted Distance Metrics K-Means (VWDM-KMeans) method that assigns attribute weights based on variance values to improve clustering quality. The proposed approach consists of Min-Max Normalization, variance calculation, weight generation, and integration of variance-based weights into the distance metric used by K-Means. Experiments were conducted on a non-linear dataset containing 103 records and 3 attributes (x, y, and z) with K = 3 clusters. The generated attribute weights were 0.3207, 0.3342, and 0.3451 for attributes x, y, and z, respectively. The performance of VWDM-KMeans was compared with conventional K-Means and K-Medoids using the number of iterations, Sum of Squared Errors (SSE), and Silhouette Score (SS). The results showed that VWDM-KMeans converged in 5 iterations, compared to 6 iterations for K-Means and 3 iterations for K-Medoids. In terms of cluster compactness, VWDM-KMeans achieved the lowest SSE value of 2.7932, outperforming K-Means (8.2429) and K-Medoids (8.9602). Furthermore, VWDM-KMeans obtained a Silhouette Score of 0.4854, equal to K-Means and higher than K-Medoids (0.4696). These findings demonstrate that incorporating variance-based attribute weighting into the distance calculation process improves cluster compactness while maintaining cluster separation quality and stability. Therefore, VWDM-KMeans can serve as an effective and computationally efficient alternative for clustering non-linear data.]]>
Copyrights © 2026
