An Intrusion Detection Approach Based on Improved Deep Belief Network and LightGBM

With the development of network technology, network attacks and intrusion methods have become more and more complex and diverse. Aiming to solve the problems of low classification accuracy and high false-positive rate (FPR) in existing intrusion detection methods, this paper proposes an intrusion detection approach based on improved Deep Belief Network (DBN) and Light Gradient Boosting Machine (LightGBM). Then, non-linear feature extraction based on the improved DBN is performed, and the new dataset obtained is given as input to a LightGBM for classification. Finally, experimental results performed on the dataset present accuracy of 97.73% and an FPR of 1.32%. The results show that the proposed approach has great improvement in classification accuracy and FPR.