To Improve the Efficiency and Accuracy of Optimal Features for Classification Using Mutual Information Based Algorithm

Abstract

Feature selection is a technique for eliminating irrelevant and redundant features and selecting the most optimal subset of features that produce a better characterization of patterns belonging to different classes. In this Paper proposes a new filter-based feature choice approach, in which theoretical evaluation of mutual statistics is brought to assess the dependence between functions and output instructions. The maximum relevant features are retained and used to assemble classifiers for respective training. As an enhancement of mutual information characteristic selection (MIFS). Redundant and beside the point features in records have triggered a protracted-time period problem in network visitors type. These functions now not simplest sluggish down the method of type but also prevent a classifier from making correct choices, especially while coping with large records. in this paper, we propose a mutual statistics primarily based algorithm that analytically selects the most reliable function for classification. These mutual records based characteristic choice set of rules can take care of linearly and nonlinearly structured facts functions.