Intrusion detection method based on constrained Boltzmann machine and delayed decision
Liang Hong, Han Bin · 2022
In the face of massive dimensional and nonlinear data, traditional intrusion detection algorithms suffer from inadequate feature extraction and inaccurate classification models. To this end, an intrusion detection method based on restricted Boltzmann machine (RBM) and delayed decision is proposed. The RBM is selected to perform feature extraction on the dataset, delayed decision is performed in the classification decision stage for behaviors that cannot be immediately classified and grouped into the boundary domain, and the feature extraction process is further repeated for behaviors in this domain and different granularity feature spaces are constructed, and finally the classification results are output. The experimental results show that the accuracy of this method on the NSL-KDD dataset is 96.1%, which is 2.5 percentage points higher than the hierarchical intrusion detection system based on spatio-temporal features, which has the highest accuracy among the compared methods.