Investigating Intrusion Detection Using a Culturally-Inspired Genetic Algorithm Trained Neural Network Ensemble
ISSN:
Description
Internet’s popularity alongside proliferation of smartphones as mode of data exchange for businesses as a great strategy for information sharing amongst users on a private company network. It has consequently also, attracted adversaries with proliferation of attacks to exploit unsuspecting users of resources for personal gain. Adversaries utilize socially-engineering attacks, to breach and gain unauthorized access to a compromised user device via subterfuge mode that can also deny such users of access to resources on a network. With denial-of-service carefully crafted to wreak havoc on network infrastructures, it has since become expedient to explore deep learning mode to predict such cases performance. We explore a scheme to effectively distinguish between genuine and malicious packets; And benchmarks our results using XGB, Random Forest, and Decision Tree. Result shows that our model yields F1 0.9945 that outperforms XGB, RF and DT with F1 of (0.9925, 0.9881 and 0.9805). Its Accuracy of 0.9984 outperforms XGB, RF, and DT with (0.9981, 0.9964 and 0.9815) respectively). Our model correctly classified 13,418 cases with 99.84% Accuracy with 283 cases incorrectly classified. Thus, model effectively differentiates genuine from malicious packets. Keywords: Intrusion. DoS, Transfer Learning, Korhonen Neural Net, Genetic Algorithm