• Growing advancement in cyber-attacks in the form of network intrusion is a major threat to organizational networks. Its efficient and timely detection has been a huge challenge to traditional intrusion detection methods owing to evolving complexity of modern Big Data Networks and increasing sophistication of cyber-attacks tactics. • Deep learning and neural network models are computationally expensive in terms of required resources, training and testing time for NIDS, especially for a resource constrained environment like the SMEs.
The aim of this work was to review the performance of current ensembles of traditional machine learning models for Network Intrusion Detection (NID) and propose an ensemble model with improved performance accuracy level as against deep learning solutions.
To research into the machine learning (ML) pipeline of ensemble of simple traditional ML models, optimize every phase towards building ensemble model with higher predictive performance accuracy over existing ones, while still benefitting from the lower computational cost, robustness and adaptability that ensemble learning of traditional ML models offers over deep learning.
Ensemble Learning, Deep Learning, Machine Learning, Stacking with Meta-learning, Bagged undersampling, Z-score & Min-Max Normalization, Mutual Information & SelectKBest Feature Selection, Naive Bayes, Random Forest, Extreme Gradient Boosting, Decision Trees, Hyperparameter Tuning with Cross-validation Settings, RandomSerachCV, Python, Python Notebook, Anaconda, R-Programming, CICIDS2017