A feature selection algorithm based on redundancy analysis and interaction weight
- PDF / 1,168,654 Bytes
- 15 Pages / 595.224 x 790.955 pts Page_size
- 17 Downloads / 166 Views
A feature selection algorithm based on redundancy analysis and interaction weight Xiangyuan Gu1 · Jichang Guo1 · Chongyi Li1 · Lijun Xiao1 Accepted: 11 September 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract The performance of some three-dimensional mutual information-based algorithms can be affected, since only relevance and interaction are considered. Aiming at solving the problem, a feature selection algorithm based on redundancy analysis and interaction weight is proposed in this paper. The proposed algorithm adopts three-way interaction information to measure the interaction among the class label and features, and processes features for interaction weight analysis. Then, it employs symmetric uncertainty to measure the relevance between features and the class label as well as the redundancy between features, and selects the features with greater relevance and interaction as well as smaller redundancy. To validate the performance, the proposed algorithm is compared with several feature selection algorithms. Since relevance, redundancy, and interaction analysis are all presented, the proposed algorithm can obtain better feature selection performance. Keywords Three-way interaction information · Symmetric uncertainty · Redundancy analysis · Feature selection
1 Introduction As an important way of dimensionality reduction, feature selection employs metrics to measure the original features and selects some features with better performance from them [1–3]. Feature selection can be applied to many fields, such as text processing [4, 5], steganalysis [6, 7], network anomaly detecting [8], and underwater objects classification [9]. Mutual information is a metric in feature selection and there are some algorithms based on mutual information, such as Mutual Information based Feature Selection (MIFS) [10] and Minimal-Redundancy-MaximalRelevance (mRMR) [11]. They employ mutual information to measure the relevance and redundancy. Since, interaction is not considered, their performance is affected. Three-dimensional mutual information is a supplement of mutual information and it includes three-way interaction
Jichang Guo
[email protected] 1
School of Electrical and Information Engineering, Tianjin University, Tianjin, 300072, China
information. Some algorithms based on three-way interaction information are proposed, such as Dynamic Weightingbased Feature Selection Algorithm (DWFS) [12] and Interaction Weight based Feature Selection Algorithm (IWFS) [13]. They adopt symmetric uncertainty to measure the relevance and exploit three-way interaction information to measure the interaction, and select the features with greater relevance and interaction. Their performance can be influenced due to ignoring redundancy. To promote the performance, relevance, redundancy and interaction are all considered, a feature selection algorithm based on redundancy analysis, and interaction weight is proposed. The algorithm initializes the weight of features, and employs symmetric uncertainty to measure th
Data Loading...
