Parameter optimization of support vector machine based on improved cuckoo search algorithm

Abstract

To solve the problem of difficult selection of penalty factor and kernel function parameters of Support Vector Machine (SVM), an improved cuckoo search algorithm (GFCS) was proposed to optimize SVM parameter model (GFCS-SVM) . The GFCS algorithm improves the optimization ability of the original cuckoo search algorithm from three aspects: replacing fixed discovery probability with dynamic discovery probability, adaptively adjusting the step size control factor of cuckoo Levy flight, and improving the dynamic inertia weight in the updating formula of cuckoo random walk. The penalty factor and kernel function parameters of SVM were optimized by GFCS algorithm, which was tested on a UCI data set, which was tested on a UCL data set compared with traditional SVM, particle swarm optimization SVM, firefly optimization SVM and cuckoo search algorithm optimization SVM, the GFCS algorithm has the highest classification accuracy and is an effective SVM parameter optimization algorithm.

Keywords

• support vector machine
• cuckoo search algorithm
• parameter selection
• adaptive step size
• dynamic discovery probability
• dynamic inertia weight