Dimensionality Reduction and Data Partitioning with Feature Hybridization Scheme

Abstract

Data mining and machine learning methods are applied to extract knowledge from large databases. Dimensionality is the key issue in the data mining and machine learning applications. The high dimensional data analysis requires huge computational resources and processing time. The performance and accuracy are reduced with reference to the irrelevant, noisy and redundant features. Dimensionality methods are applied for better visualization, data compression, noise removal, understandability and generalization factors. Text mining, web mining, image processing and bioinformatics applications are build with dimensionality reduction methods. Dimensionality reduction is carried out with two models Feature Selection (FS) and Feature Extraction (FE). Feature selection discovers the suitable features from the original set of features. The feature extraction method transforms the original set of features into required form. The compound feature generation (CFG) model integrates the feature selection and extraction methods to fetch the original and transformed features. The Minimum Projection error Minimum Redundancy (MPeMR) framework is build with Unified iterative algorithm to fetch features in supervised and unsupervised cases. The Compound Feature generation (CFG) method is build with pairs of features in minimum projection error and redundancy estimation process. The feature hybridization scheme is build to combine the original and transformed features with generalized matching criteria. The feature integration operation is improved with diverse feature count based models. The data partitioning process is carried out on the dimensionality reduced data with K-Means clustering algorithm.