An Intrusion Detection Method Using ADASYN and Bayesian Optimized LightGBM

To address the problems of traditional network intrusion detection models with long training and detection time, difficult hyperparameter selection, and poor detection performance due to data imbalance, this paper proposes an intrusion detection method named AB-LightGBM, which utilizes adaptive synthesis (ADASYN) oversampling technique and Bayesian optimized light gradient boosting machine (LightGBM). In order to solve the problem of poor detection rate of a small number of attacks caused by imbalanced training data, the ADASYN algorithm is used to increase training samples. In addition, to increase accuracy and reduce computing burden, Bayesian optimization is employed to tune the hyperparameters of the LightGBM model. Validated on the NSL-KDD and CICIIDS2017 data sets, the accuracy rates are as high as 95.46% and 99.89%, respectively. The method proposed in this paper outperforms other recent methods in terms of accuracy and false alarm rate.