Data Mining and Knowledge Engineering

Abstract

The conventional clustering algorithms similar as K-means and probabilistic clustering, their clustering outcomes are complex to the existence of outliers in the records. Uniform a few outliers can transfer the bulk of these algorithms to detect significant hidden structures representation their outcome are unpredictable. In this mounting robust clustering algorithms that intention to cluster the data, as well as to detect the outliers. Proposed new method depends on the infrequent existence of outliers in the data, which clarifies to sparsity in a carefully preferred area. Leveraging sparsity in the outlier domain, outlier-responsive robust K-means and probabilistic clustering approaches are projected. Their novelty lies on recognizing outliers while carrying out sparsity in the outlier domain complete sensibly preferred regularization. A block co-ordinate descent method is established to attain iterative algorithms with conjunction promises and minor additional computational density relating to their non-robust matching part. Kernelized versions of the robust clustering algorithms are also established to capably hold high dimensional data, to detect nonlinearly divisible clusters, or even cluster objects that are not characterized by vectors. Hence experimental results are built on the polynomial kernel for outlier.

Keywords

Full Text