关键词: 半监督SVM, 遗传FCM, 样本预选取

Abstract: Aiming at traditional semi-supervised support vector machine training methods spend most of time on non support vectors，we propose the pre-selection sample method of genetic algorithm fuzzy C mean for working dataset.In the optimum learning process，working dataset（unlabeled data） is added to original training dataset（label data） to construct new training dataset.Firstly the new method is to divide unlabeled data（working datasets） into many subsets with a new label by Genetic Fuzzy C Means clustering，then train the concave SVM using the new data set to get decision boundary and support vectors，at last use the SVM classifier to classify the unlabeled data.The method proposed can select boundary vectors which have most probability to be support vectors to training set and save memory through decreasing working dataset.When working dataset is decrease to some scale of original working dataset，experimental results of three dimensional random data show that both classification accuracy rate and number of support vectors between reducing working dataset and original working dataset is almost same.But classification time of reducing working dataset is largely decreasing.It is illustrated that the result of sample pre-selection is satisfied.

Key words: semi-supervised Support Vector Machine, Genetic algorithm Fuzzy C Mean, sample pre-selection