An Improved Fish Swarm Algorithm for Neighborhood Rough Set Reduction and its Application

In this paper, an improved fish swarm algorithm for neighborhood rough set reduction (IFSANRSR) is proposed. In IFSANRSR, by introducing an adaptive function to control the visual and step size of artificial fish, the problem of inconsistent convergence speed existed in a traditional artificial fish swarm algorithm (FSA) is avoided. The movement of artificial fish in the swarming and following behavior is improved to shorten the running time of the algorithm. The searching behavior is improved to enhance the local search ability without changing the global searching ability of the algorithm. By introducing the mechanism of extinction and rebirth, the worst solution is eliminated and rebirth takes place after each iteration, which ensures a high level of the overall fitness. The experimental results on three datasets from the University of California at Irvine (UCI) show that the attributes reduction by using the IFSANRSR has higher reduction rate and classification accuracy in most cases. It could better deal with real-valued data attributes and ensure the optimal attribute reduction set to be found. The experimental result on the decision system of aluminum alloy welded joints show that by using the IFSANRSR, the key influencing factors of fatigue life of aluminum alloy welded joints could be obtained, and the weight of each influencing factor could be calculated quantitatively.