A PCA + RF-Vl Hybrid Approach for feature extraction in Intrusion Detection Systems
C Mallaradhya, G N K Suresh Babu · 2024
Intrusion Detection Systems (IDSs) are essential components of network security, tasked with identifying and thwarting potential cyber threats. However, contemporary IDSs encounter several hurdles that impede their effectiveness. These challenges include grappling with redundant data characteristics, detecting intricate lateral movement cyberattacks, and attaining optimal classification accuracy. To surmount these obstacles, this study advocates for a novel approach that merges Principal Component Analysis (PCA) with Random Forest version I (RF -VI) for feature extraction and classification within IDS frameworks. The proposed PCA+RF-V1 strategy amalgamates the strengths of PCA in dimensionality reduction and RF -VI in intruder feature extraction, culminating in a comprehensive enhancement of IDS performance. Extensive evaluations conducted on benchmark datasets such as CICIDS2017 provide empirical evidence of the hybrid approach's efficacy, assessed through metrics like Fl-score, precision, recall and accuracy. The results underscore the proficiency of PCA in streamlining data dimensions while retaining critical information, thereby empowering RF -VI to extract pivotal features essential for precise intrusion detection. Notably, the PCA+RF-VI hybrid model achieves remarkable performance metrics, including 97.8% accuracy, 98.2% precision, 97.5% recall, and 97.8%FI-score. These findings illuminate the potential of the PCA+RF-VI hybrid paradigm in mitigating the evolving cyber threat landscape. Leveraging PCA's dimensionality reduction capabilities and RF - VI's robust feature extraction mechanisms, the proposed approach presents a promising avenue for fortifying cyber security defenses and fortifying network infrastructure resilience against malicious infiltrations.