Neighborhood Preprocessing SVM for Large-Scale Data Sets Classification

Support vector machine (SVM) has been a promising method for data mining and machine learning in recent years. However, the training complexity of SVM is highly dependent on the size of a data set. A preprocessing support vector machines (P-SVM) method for large-scale data set classification is presented to speed up SVM training. By analyzing the neighbor classification feature for each sample in training data set, a decision criterion was built to keep or delete this sample from the original data set without losing the classification. The new method can provide an SVM with high quality samples. Experiments with random data and UCI databases show that SVM with our new preprocessing method retains the high quality of training data set and the classification accuracy very well.