Abstract:Intrusion detection systems are essential to protect the network security. However, it is hard for a traditional single algorithm to attain satisfied detection results for different attack classes. To solve this problem, this paper proposes an intrusion detection method based on Extreme Learning Machine (ELM) and modified K-means. ELM algorithm is optimized by Parametric Rectified Linear Unit (PReLU) activation function. The modified K-means algorithm can automatically select the initial centroids of clusters and the number of clusters by setting the distance threshold. Based on cascade algorithms, a hybrid intrusion detection method is designed based on improved ELM and modified K-means. The experimental results on NSL-KDD dataset show that, compared with other traditional algorithms such as BP neural network, Support Vector Machine (SVM) and ELM, the proposed method improves the detection results and reduces the false alarm rate.