Research On Twin Support Vector Machines And Its Optimization Methods

Twin Support Vector Machines (TWSVM) is a new machines learning methodbased on the theory of Support Vector Machines (SVM). Unlike SVM, TWSVMgenerates two nonparallel hyper-planes by solving two smaller-sized QPPs such thateach hyper-plane is closer to one class and as far as possible from the other. Thestrategy of solving two smaller-sized QPPs, rather than a single large one, makes thelearning speed of TWSVM approximately four times faster than that of the standardof SVM. At present, TWSVM has become one of the popular methods because of itslow computational complexity. However, the efficiency of TWSVM still can beimproved, and the parameters should be optimized, this paper made a systematicstudy to solve these issues, and the main contents are as follows:Since TWSVM is proposed, many scholars improved this algorithm and furtherproposed least squares twin support vector machines, projection twin support vectormachines and so on and applied it in biomedicine, speech recognition and other fields.After in-depth study of the basic ideas and algorithm process of TWSVM, and inorder to further improve the classification efficiency, we consider to optimize it bypretreatment and propose Twin Support Vector Machines based on Rough Sets(RS-TWSVM), this algorithm uses the rough sets theory to reduce the attributes. Thefinal experimental results and data analysis verify the feasibility and effectiveness ofthe algorithm.In the research and experiments of TWSVM and RS-TWSVM, we find that it isdifficult to specify the parameters of them, so we continue to explore how to improveTWSVM by parameter optimization, and then we propose the twin support vectormachines based on particle swarm optimization (PSO-TWSVM). This algorithm usePSO to find the parameters for TWSVM, so that blindly parameters selection isavoided. The experimental results show that this algorithm is effective.In order to explore the pros and cons of optimize parameters of TWSVM bydifferent swarm intelligence optimization algorithms, we propose Twin SupportVector Machines based on Fruit Fly Optimization Algorithm (FOA-TWSVM), thisalgorithm uses the latest swarm optimization algorithm--Fruit Fly OptimizationAlgorithm to optimize the parameters of TWSVM, and it can also avoid to choiceparameters blindly. Finally, the experimentation in MATLAB verifies FOA-TWSVMis feasible. Then we compare and analysis the advantages and disadvantages of two optimization methods.