Optimization of predictive performance of intrusion detection system using hybrid ensemble model for secure systems

Area of Science:

• Cybersecurity
• Machine Learning
• Network Security

Background:

• Network intrusion poses a significant threat to organizational security.
• Small and medium enterprises (SMEs) in developing countries face resource limitations in implementing effective security solutions.
• There is a need for robust, cost-effective intrusion detection systems (IDS) for both traditional and Internet of Things (IoT) networks.

Purpose of the Study:

• To propose a novel hybrid ensemble model for enhancing the predictive performance of intrusion detection systems (IDS).
• To develop a cost-effective and efficient IDS suitable for SMEs.

Main Methods:

• A hybrid ensemble model combining Random Forest and Recursive Feature Elimination (RF-RFE) was developed.
• The model was evaluated on benchmark datasets: NSL-KDD, UNSW-NB15, and CSE-CIC-IDS2018.
• Performance was compared against deep learning paradigms, focusing on accuracy, computational cost, and training time.

Main Results:

• The proposed RF-RFE ensemble model achieved high overall accuracy: 99% on NSL-KDD, 98.53% on UNSW-NB15, and 99.9% on CSE-CIC-IDS2018.
• The machine learning ensemble method demonstrated state-of-the-art results.
• The model achieved these results with lower computational cost and less training time compared to deep learning approaches.

Conclusions:

• The proposed hybrid ensemble model significantly optimizes the performance of intrusion detection systems.
• This research contributes to developing more secure systems and applications, particularly beneficial for resource-constrained SMEs.
• The findings highlight the effectiveness of machine learning ensemble methods in addressing network security challenges.