A DDoS Detection Method with Feature Set Dimension Reduction

With the advent of fifth-generation network, mobile internet security suffer plenty of DDoS attacks. The number and frequency of occurrence of DDoS attacks are predicted to soar as time goes by, hence there is a need for a sophisticated DDoS detection framework to 5G network without worrying about the security issues and threats. Normally, the neural networks are widely used to detect complex and diversified DDoS attacks. However, feature vectors with high dimensions have a negative effect on detection performance. At present, there is little work on DDoS security dataset dimensionality reduction and verification. This paper proposes a DDoS detection method based on dimensionality reduction security dataset. First, XGBoost and mutual information algorithms are used to reduce the dimensionality of the KDDCup99 and CICDDoS2019 dataset respectively. Futhermore, we collect dataset in the experimental environment. Then, the CNN+LSTM and MLP neural network detectors are used to detect the dataset before and after the XGBoost dimensionality reduction. The experimental results show that using the XGBoost dimensionality reduction dataset, the neural network detector can detect multiclassify DDoS attack types with high accuracy and recall rate.