A kind of SVM fast training method based on samples segmentation learning

摘要

Aiming at the problem of slow training in case of Support Vector Machines (SVM) training in massive samples, the paper analyzed relationships and rules between training samples number and training time. The good characteristic of small samples learning of SVM was utilized to lay basis for SVM fast algorithm based on samples segmentation learning, which divided massive samples into small partition for training and obtained classifier after weighted processing on multiple SVM. As proved by experiment, the method can greatly improve training speed, while ensuring good generalization at the same time.