An Efficient Hybrid Feature Selection model for Dimensionality Reduction

摘要

This paper presents a novel approach based on hybrid feature selection that significantly reduces dimensionality of features. In this paper, an efficient method consisting of ReliefF and PCA is proposed which shows remarkable results with different chronic disease datasets. The presented work is suitable for both text and micro-array datasets which determines the optimal value of threshold for the selection of relevant and non-redundant features. To validate the performance of proposed work, ten popular benchmark datasets are used. With the results obtained, it is found that the presented approach reduces more than 50% irrelevant and redundant features from the dataset. Also with the proposed method, the computation time significantly decreases for all considered chronic disease datasets. Moreover, it is experimentally depicted that the threshold value significantly affects the selection of appropriate features.