A Novel Binary Black Hole Algorithm for Feature Selection of Scientific and Technical Text
Xian Wei Wu, Kaixiang Yi, Minrui Fei, Wenju Zhou, Chenhui Li, Yayu Chen · 2021
Selecting effective features from the raw attributes is an important step in scientific and technical text classification. Feature selection aims to pick the best subset of candidate attributes to achieve superior performance of the learning algorithm. Recently, many meta-heuristic algorithms were widely developed for feature selection in classification problems. Black Hole Algorithm (BHA) is an emerging meta-heuristic algorithm, inspired by the mechanism of star motion around black holes, which has achieved excellent performance on a variety of optimization problems. A novel Binary Black Hole Algorithm (BinBHA) was proposed in this work, to enhance the efficiency of searching optimal features from the high-dimensional attributes of scientific and technical texts. In BinBHA, all operators are binary encoded without continuous-binary transformation, which makes it better for exploring the solution space of discrete problems and finding globally optimal solutions. The proposed algorithm has been compared with several alternative state-of-art methods for feature selection, and the effectiveness of the approach was evaluated on several benchmark datasets. In this paper, BinBHA was successfully implemented on an actual scientific and technical text dataset, and achieved the best performance results among other algorithms. The experimental results demonstrated that the proposed binary encoding method can improve the prediction accuracy significantly while finding the optimal solution effectively.