NMF based sparse Cholesky decomposition technique for dimensionality scale back in large data sets

Abstract

This paper aims at resolving the issues related to increased dimensionality of data in data mining. In this paper, Sparse Cholesky decomposition (SCD) is combined with Non-integer Matrix Factorization (NMF) to remove the problems arising due to increased data dimensionality. The increased data dimensionality in datasets is probably due to non-orthogonality of datasets. The complex conjugate values is used to remove the sparse matrix and a complex gradient algorithm reduces the sparse matrix by the extraction of conjugate values. The SCD-MNF extracts the feature vector and upper triangular matrix linearly maps the feature vector obtained from the SCD. Hence, NMF is employed with SCD for structuring the datasets and this helps to form a well-defined data geometry. The proposed system is evaluated against normalized mutual information and accuracy against different text datasets. The results prove that SCD-NMF attains better results than conventional methods in finding the instances related to the given query.