A method for improving packet detection accuracy using artificial neural networks in computer networks

This paper proposes an effective method for detecting malicious network data by combining sparse-response deep belief network (SR-DBN) and support vector machine (SVM). SR-DBN is an efficient unsupervised learning model that extracts redundant-free data feature representations, while SVM is adopted to construct a classifier with strong generalization capability in the feature space through supervised training. In this method, SR-DBN is used to realize feature representation of abnormal network payloads, and SVM is applied to complete the classification of normal and abnormal payloads.
Simulation and experimental results demonstrate that the proposed network anomaly detection system achieves a higher detection rate than the stacked autoencoder-based multilayer perceptron.