Based on Improved Semi-Supervise Clustering Method Training Classifier for Analog Circuit Fault Classification

In recent years, semi-supervised clustering as an important research subject has significance in dealing with lack of training sample sets. However, formerly semi-supervised clustering usually cannot attend satisfactory consequence in precision and training time at the same time. Aimed to the problem of clustering method assist training classifier to label the samples, produce the time optimization algorithm. Based on prior knowledge, mining the acquired unlabeled sample sets deeply of their potential data structure and combine semi-supervised fuzzy C-means(SS-FCM) arithmetic with similarity coefficient to sort out the samples for training time improvement. On the basis of little influence on classification result accuracy, gain the fuzzy similarity matrix from Euclidean distance and assess the maximum dependable sample point with its neighborhood for their similarity degree, will avoid searching the maximum dependable sample point one by one and optimize holistic clustering time costing from reduce the iterations of classifier to some extent. Through artificial circuit simulation experiment, using improvement SS-FCM assist SVM classifier and single SVM and SS-FCM assist SVM classifier to make a comparison, verify the algorithm from classify precision and arithmetic speed and the result of experiment can prove the validity of the improvement.