PROJECT TITLE :

Feature Selection by Maximizing Independent Classification Information - 2017

ABSTRACT:

Feature selection approaches primarily based on mutual information will be roughly categorized into 2 teams. The first group minimizes the redundancy of features between each alternative. The second cluster maximizes the new classification information of features providing for the chosen subset. A important issue is that enormous new data will not signify little redundancy, and vice versa. Features with giant new data but with high redundancy may be selected by the second cluster, and options with low redundancy but with little relevance with categories might be highly scored by the first cluster. Existing approaches fail to balance the importance of both terms. As such, a replacement info term denoted as Independent Classification Data is proposed during this paper. It assembles the newly provided information and the preserved data negatively correlated with the redundant info. Redundancy and new data are properly unified and equally treated in the new term. This strategy helps realize the predictive features providing massive new info and little redundancy. Moreover, freelance classification information is proved as a loose upper sure of the entire classification data of feature subset. Its maximization is conducive to realize a high global discriminative performance. Comprehensive experiments demonstrate the effectiveness of the new approach.