Applying Ensemble Deep Learning for Enhanced Security in Platform Engineering

Abstract

Cyberattacks are becoming increasingly complicated, due to which it is getting harder to reliably identify intrusions. Inability to eradicate any intrusion might harm the reputation of security of services, such as data availability, privacy, and sincerity. This research provides a deep ensemble intrusion detection system (IDS), which integrates various deep learning models that consist a hybrid feature selection mechanism, to overcome the problems of low detection accuracy and high false-positive rate. A hybrid feature selection method that includes genetic algorithm particle swarm optimization and ant colony optimization is used to decrease the number of features in the training datasets. Then, a three-stage meta-classifier ensemble formed by a bidirectional long short-term memory, a convolution neural network, and a random forest is proposed. Finally, a voting classifier makes the final decision of classification based oh majority voting. Experiments using CICIDS-2017, CSE-CICIDS-2018, and UNSW-NB15 datasets validated the suggested network IDS that surpasses the existing methods. The classification accuracy of our method proposed was found to be 98.31%, 97.27%, and 98.56% for CICIDS-2017, CSE-CICIDS-2018, and UNSW-NB15 datasets, respectively.