Multi-view neutrosophic c-means clustering algorithms

Abstract

Multi-view clustering has become increasingly pervasive and prominent as multiple sources often provide different representations of information. However, existing multi-view clustering algorithms still encounter challenges since most multi-view data do not exhibit clear cluster boundaries, meaning cluster boundaries may locally overlap. Consequently, effectively characterizing and unveiling the imprecise and uncertain cluster structures in multi-view clustering remains an unresolved issue. Inspired by the robust capabilities of neutrosophic clustering in modeling imprecise and uncertain information, this paper introduces two novel multi-view neutrosophic $c$-means clustering algorithms, which can be regarded as derivatives of NCM in multi-view scenarios. The proposed algorithms are designed to represent the imprecision and uncertainty in cluster assignment of multi-view data while also autonomously discerning the importance of each view to boost clustering performance. We craft two objective functions and develop the corresponding optimization strategies to derive the neutrosophic partition matrix, view weight vector, and cluster centers matrix. Through extensive testing on both synthetic and real-world datasets, we demonstrate the practicality and effectiveness of our proposed algorithms.