Optimized feature selection algorithm for high dimensional data

Abstract

Objectives: This research paper, based on fuzzy entropy, adapts a new method along with firefly concept, seeks to select quality features. At the same time it removes redundant and irrelevant attributes in high dimensional data. Methods/Statistical Analysis: Feature selection can be understood as a data prepossessing method in order to reduce dimensionality, eliminate irrelevant data and sharpening of accuracy. In the pattern space, fuzzy entropy is used to estimate the knowledge of pattern distribution. The study of the lightning quality of the fireflies has led to the introduction of the Firefly Algorithm for computing models. This work proposes an algorithm for selecting features by integrating fuzzy entropy and firefly algorithm. Our proposed algorithm's performances are analyzed using four different high dimensional data sets WILT, ORL, LC and LTG. Findings: The algorithm which is introduced here is further experimented with four variant data sets and the results shows that this algorithm out performs the traditional feature selection method. Also our proposed algorithm achieves maximum relevance and minimum level of redundancy. The performance metrics such as sensitivity, specificity and accuracy gives significant improvement when compared with existing FCBF algorithm. Applications/Improvements: Our optimized proposed algorithm efficiently improves the performance by eliminating redundant, noisy and insignificant features and can be applied on all high dimensional data sets.