A Novel LAS-Relief Feature Selection Algorithm for Enhancing Classification Accuracy in Data mining

Abstract

Feature selection is an important task in data mining and machine learning domain. The main objective of feature selection is to find a relevant feature that predicts the knowledge better than the original set of features. This can be achieved by removing irrelevant or redundant features from original data sets. Feature selection involves a significant task of selecting relevant features from the feature space for data mining and pattern recognition. In this paper, the new approach has been introduced on feature selection on Relief based on Median Variance model. The new approach is named as LAS-Relief algorithm. This algorithm facilitates to stabilise the feature weights estimation compared to mean variance based Relief algorithm and is considered to be a better successful algorithm for feature selection. The random selection of instances in the data sets will lead to the fluctuation of weight estimation. This in turn leads to poor evaluation accuracy. This new approach removes the irrelevant features in the feature space. The novel LAS-Relief algorithm incorporates the median variance in the feature weight estimation. The feature weight is calculated by selection of the instances in random.  To overcome this issue, the novel feature selection algorithm called LAS-Relief algorithm is proposed based on median variance. This algorithm takes both the median and the variance of difference between instances. These are considered as the criterion of feature weight estimation in this LAS Relief algorithm.. This algorithm makes the result more stable and more accurate on classification. The relevant features are obtained from the original feature space using LAS-Relief algorithm, which outperforms well than Mean Variance Relief algorithm.