Application of SVM in anomaly detection based on sampling and feature extraction

In order to identify the four types of attacks in the KDD99 data set, an anomaly detection model based on supporting vector machines is proposed. The model uses supporting vector machines to train five classifiers and detects the attacks by integrating the classification results of the five classifiers. Aiming at the problem that the detection model has a low recall rate when detecting some categories, a feature extraction method based on abnormal proportion analysis is proposed to extract effective feature combinations for each classifier, and a sampling technique is used to preprocess the training set of each classifier. Experimental results show that the improved anomaly detection model can not only improve the performance of each classifier, but also improve the comprehensive discrimination performance of the five classifiers.