A Semi-supervised Intrusion Detection Algorithm Based on Natural Neighbor

Training samples of the intrusion detection algorithms based on supervised learning is hard to acquire. The accuracy of the intrusion detection algorithms based on unsupervised learning is low. Common semi-supervised intrusion detection algorithms need parameter k which is selected by human. To solve these problems, a semi-supervised intrusion detection algorithm based on natural neighbor is proposed. Natural neighbor (2N) proposed by us is a novel concept on nearest neighbor. It does not need parameter k when search neighbors of each point. The specific steps of the intrusion detection algorithm are as follows: first, do clustering based on 2N on labeled data. Then, make classification based on 2N on unlabeled data according to the result of clustering. The experimental result shows that the algorithm works well both in detection accuracy and stability.