Deep Belief Network Integrating Improved Kernel-Based Extreme Learning Machine for Network Intrusion Detection

Deep learning has become a research hotspot in the field of network intrusion detection. In order to further improve detection accuracy and performance, we proposed an model based on improved deep belief (DBN). Traditional neural training methods, like Back Propagation (BP), start train with preset parameters such as randomly initialized weights thresholds, which may bring some issues, e.g., attracting local optimal solutions, or requiring long period. We use Kernel-based Extreme Learning Machine (KELM) supervised ability replace BP algorithm DBN bid ameliorate situation. Considering problem poor classification performance usually caused by initializing kernel KELM, enhanced grey wolf optimizer (EGWO) is designed optimize KELM. search optimization traditional algorithm, novel strategy combining inner outer hunting introduced. Experiments KDDCup99, NSL-KDD, UNSW-NB15 CICIDS2017 datasets show that DBN-EGWO-KELM greater advantages terms its accuracy, precision, true positive rate, false rate other evaluation indices compared BP, RBF, SVM, LIBSVM, CNN, DBN-KELM models, can effectively meet requirements complex networks.