A boosted hierarchical clustering linkage algorithm: K-Centroid link supported with OWA approach

The choice of linkage algorithm plays a crucial role in determining the quality of hierarchical clustering and therefore must be made carefully. This selection significantly influences the effectiveness of the clustering process. However, conventional linkage methods do not take into account the influence of records located near the cluster centers. Previous studies proposed the k-centroid link, a new cluster merging criterion that analyzes instances near cluster centers in greater detail to improve clustering quality. The k-centroid link computes the average distance among the k nearest data points to the central point within each cluster. In this study, we enhance the clustering capability of the k-centroid link by integrating the Ordered Weighted Averaging (OWA) approach. Specifically, OWA values of the average distances between the k nearest records to each cluster center are calculated using a constant-level weighted stress function across different α values, rather than relying solely on direct distance calculations. The proposed model was evaluated on 24 publicly available benchmark datasets specifically designed for clustering tasks. The results demonstrate that the k-centroid link can be significantly improved through the application of OWA-based approaches with different stress functions.