Support vector machine based on new fuzzy membership

The definition of fuzzy membership is the key in fuzzy support vector machine, current methods only are considered the efficient of outliers and noises which are far away from the center of the training set. If outliers and noises are close to its cluster center, the fuzzy membership calculated by current method will be unreasonable, sometimes even ridicules. A new fuzzy membership was proposed, which considered both position of far and near noises in the training set to cluster center, and affinity among samples are included simultaneously. Results of simulation show that the algorithm can get better separating hyper-planes. Experimental results show that the algorithm is more estimating accuracy than other related algorithms.