Evidential Clustering with Soft Pairwise Constraints

Constrained clustering, a prominent field in data analysis, integrates prior knowledge or constraints into the clustering process. This paper introduces a novel extension to the Evidential C-means (ECM) algorithm, incorporating soft pairwise constraints to enhance clustering performance in scenarios with available prior knowledge. The proposed method significantly augments the algorithm's capabilities, providing a more nuanced approach to modeling relationships between data points through soft constraints. Unlike traditional hard constraints, the introduced soft constraints accommodate uncertainty, enabling a realistic representation of complex data relationships. Additionally, an adaptive metric overcomes the limitations of the Euclidean distance, adept at handling non-spherical classes. Experimental results demonstrate the effectiveness of these extensions, showcasing superior performance compared to other constraint-based methods while requiring fewer constraints for accurate clustering outcomes. The applicability of the proposed method is demonstrated through experiments on real datasets, including a medical image segmentation application.