4th International Conference on Intelligent and Fuzzy Systems (INFUS), Bornova, Turkey, 19 - 21 July 2022, vol.504, pp.142-150
Sparse clustering algorithms create clusters with a sparsity of features and are effectively applied to extremely high-dimensional single-view datasets containing irrelevant features, noise data, outliers, and missing data. Multi-view datasets contain features extracted from various measuring devices using the same sample. Each distinct group of features constitutes a specific view of the data. Although these data are extracted from diverse settings and domains, they must be highly correlated since they are from the same sample. Most of the sparse clustering methods can only handle single view datasets. Thus, they may encounter difficulty obtaining the desired result when clustering multi-view datasets, leading to poor performance. Therefore, there is a need to develop a suitable multi-view clustering method to address the above problem. In this paper, we propose an algorithm called Sparse Weighted Multi-view Possibilistic C-means with L1 Regularization (S-WMV-PCM-L1) designed to perform multi-view clustering as well as view and feature selection with Possibilistic C-Means (PCM) as its base function and using L1 (LASSO) Regularization as a penalty term. This algorithm uses a weighting scheme within its clustering framework to determine the comparative significance of each view of data points and features. Experimental results using real-world datasets show the feasibility and effectiveness of our proposed algorithm. Also, S -WMV-PCM-L1 performs better when compared to other existing multi-view clustering algorithms.