Abstract: A fuzzy support vector machine(FSVM)reduces the bad effect of outliers or noises on the classifier by using different training parameters for different training examples．This paper proposed an FSVM based on pre-classification,in which fuzzy weight for each training example is computed according to a classification hyperplane created for the training example．The method considers not only the effect of each individual training example,but also the sensitivity of the classifier to the outliers or noises．SVM with such fuzzy weights is able to make proper suppression for the training data．Experimental results on the standard training datasets from IDA and UCI repositories show the reasonableness and effectiveness of the proposed algorithm.

Key words: Fuzzy support vector machine(FSVM),Pre-classification hyperplane,Fuzzy weight,Sensitivity