Self-adaptive attribute weighted neutrosophic c-means clustering for biomedical applications

Abstract

The applications of clustering in biomedical is pervasive and ubiquitous. A typical example is gene expression data analysis, where clustering is emerging as a powerful solution for uncovering cancer-related insights. Neutrosophic c-means (NCM) clustering has advantages over other conventional clustering methods in characterizing the uncertainty and imprecision caused by cluster overlap and identifying outliers. Nonetheless, NCM and its derivatives equally treat the contribution of each attribute to the cluster. In biomedical applications, genes (i.e. attributes) should take different importance in identifying different clusters. In this paper, we first propose a self-adaptive attribute-weighted neutrosophic c-means (AWNCM) clustering method to overcome the above defects. Moreover, a new objective function is designed to obtain optimal neutrosophic partition, cluster centers and attribute weights. Since AWNCM tends to be more effective against spherical data, we further develop a kernelized version of AWNCM, called KAWNCM, in order to better satisfy the clustering of some complex data (i.e. non-spherical data). We employ the iterative optimization strategy to obtain the optimal solutions for AWNCM and KAWNCM. The advantage of AWNCM and KAWNCM is to improve performance by learning the importance of each attribute to the cluster while maintaining an efficient solution to cluster overlap and outliers. Extensive experimental results using synthetic data and gene expression data demonstrate the feasibility and effectiveness of the proposed methods.

Keywords

• Clustering
• Neutrosophic c-means
• Gene expression data
• Attribute weighted
• Neutrosophic partition