K-Means algorithm based on multi-feature-induced order

The K-Means algorithm is a powerful tool for data analysis, but it faces several challenges when dealing with large multi-feature data. Centroid initialization and centroid determination are two significant hurdles that can reduce the performance of the K-Means algorithm. To address these challenges, based on partial-order relations, an enhanced K-Means algorithm, the multi-feature induced order K-Means algorithm (OWAK-Means) is developed which combines with a novel centroid initialization based on partial-order relations and a multi-feature induced ordered weighted average (MFIOWA) operator. By using a weighted iteration method based on partial-order relations, the OWAK-Means algorithm initializes centroids with greater precision. The MFIOWA operator is designed based on database indexing theory and the Sigmoid weight function that improves its information filtering ability. These techniques, combined with an ordered weighted distance metric and the MFIOWA operator, make the OWAK-Means algorithm an effective tool for multi-feature data analysis. In comparative analysis with the variants of the K-Means algorithm, the OWAK-Means algorithm has significant improvement in the adjusted rand score, normalized mutual information, and purity. Statistical tests, comprehensive evaluation methods, and sensitivity analysis prove that the OWAK-Means algorithm is effective and reliable.