Network Intrusion Anomaly Detection Model Based on Multiclassifier Fusion Technology

With the increasing development of industrial Internet, network security has attracted more and attention. Among numerous technologies, anomaly detection technology based on traffic become an important research field. At present, a large number methods for have been proposed. Most better performance are supervised machine learning algorithms, which require labelled data model training. However, in real network, it is impossible to manually filter label large-scale data. Network administrators can only use unsupervised algorithms actual detection, effects much worse than algorithms. To improve accuracy methods, this study proposes multiple classifier fusion technology, applies different techniques (such as Majority Vote, Weighted Naive Bayes) fuse results five best performing Comparative experiments carried out three public datasets. Experimental show that, terms RECALL AUC score, proposed achieves separate baseline robustness stability, be effectively applied wide range scenarios.