CANONICAL VARIATE FEATURE SELECTION BASED ADAPTIVE ENHANCED WINNOW MAP REDUCE CLASSIFICATION FOR PREDICTIVE ANALYTICS

Abstract

Classification is a key issue to be resolved in data mining. Few research works have been designed for performing predictive analysis through classifying the information on data warehouse. But, classification accuracy (CA) of conventional works was lower when considering a big size of data as input. In order to addresses this drawback, a Canonical Variate Feature Selection based Adaptive Enhanced Winnow Map Reduce Classification (CVFS-AEWMRC) Method is proposed. The CVFS-AEWMRC Method is designed for organizing and classifying the collected and stored data for decision making. Initially, Canonical Variate Feature Selection (CVFS) is carried out in CVFS-AEWMRC Method to select the relevant features for performing the classification. Canonical Variates analysis is a machine learning technique used to find linear combinations of features which have maximum correlation with each other. The features with maximum correlation are selected for performing the classification. Then, Adaptive Enhanced Winnow Map Reduce Classification (AEWMRC) Process is carried out in CVFS-AEWMRC Method to classify the stored data for taking decision. Adaptive Enhanced Winnow technique learns the linear classifier from labeled data samples. Winnow employs the multiplicative scheme for performing the classification process. Winnow learns the hyperplane to classify the data points for decision making. By this way, the data classification is carried out in accurate manner for decision making during the predictive analytics process. Experimental analysis of CVFS-AEWMRC Method is performed on metrics namely feature selection rate (FSR), CA, classification time (CT) and False positive rate (FPR) with number of features and data points.