A Binary-Classification Method Based on Dictionary Learning and ADMM for Network Intrusion Detection

With the rapid development of computer networks, network security becomes the focus of attention. Intrusion detection plays an important role in network security. Recently, many typical methods in machine learning have been applied to intrusion detection system, because intrusion detection can be formalized as a binary-classification issue. However, they have a strict requirement for the distribution of dataset, which need a small and balanced dataset with less noise. Few new initiatives have been proposed to handle large and imbalanced datasets. Therefore, based on dictionary learning, we proposed a novel approach called ADM-DL. With the help of alternating direction multipliers method (ADMM) algorithm, the training time of our dictionary gets shorter and the accuracy of the dictionary becomes higher. Moreover, sparse representation and the minimum principle of reconstruction error are adopted to attain a more efficient binary-classification model. ADM-DL not only reduces the complexity of processing intractable datasets, but also obtains a low-complex and high-efficient classification model. The popular KDD-CUP-1999 datasets are adopted to evaluate the performance of our proposal. The experiment results show that ADM-DL can reduce the dimension of network security data, enhance the detection rate and decrease the false alarm rate of intrusion detection.