Abstract:

Starting from the one-dimensional cellular automata model, the spread mechanism is introduced to binary glowworm swarm optimization (BGSO), and a spread binary glowworm swarm optimization (SBGSO) is proposed. In SBGSO, nutrition value and nutrition threshold is involved to each glowworm, then the spread operation is performed to produce offspring by using the method of normal distribution. Additionally, the individuals who continue to perform poorly are eliminated. The aforementioned operations can largely keep the diversity of the whole populations. After that, SBGSO is combined with rough set （RS） to handle the attribute reduction problem. When dealing with the attribute reduction problem, SBGSO is taken as a kind of search strategy and RS is taken as the evaluation criteria for attribute subsets. To analyze the feasibility and effectiveness of the proposed method, five UCI datasets are used to conduct experiments. Moreover, the 10-fold and SVM are involved to analyze the classification accuracy, experimental results show that the method has relatively higher reduction rate compared with other methods.