Network Intrusion Detection Model Based on WLA Optimized Hybrid Kernel LSSVM

To address the problems of high time overhead, low precision and poor generalization of network intrusion detection caused by high dimensionality and large differences in distribution of network data, the Hybrid intrusion detection model（HIDM） was proposed. Firstly, mutual information theory was selected as the feature selection module of HIDM model by comparing the detection effect, which realized feature dimensionality reduction and cost saving; Then the whale lifting algorithm was proposed by using nonlinear decreasing factor, adaptive weight strategy and whale optimization algorithm; Finally, the HIDM model was constructed by using the parameters of its optimized hybrid kernel least squares support vector machine, which could effectively detect network intrusions. The simulation results based on NSL-KDD dataset in the study show that the detection rate, accuracy and false positive rate of HIDM model against network attacks reach 99.63%, 99.4% and 0.86% respectively. Compared with some existing studies, the detection rate has increased; At the same time, the CICIDS2018 dataset is used to verify the generalization of HIDM model.