Speeding Up ML-Based IDSs Through Data Preprocessing Techniques

Most of the current ML-based IDSs models priotize detection accuracy over detection latency, which is critical for real-time detection and mitigation of cyber-attacks.The study evaluated the impact of Principal Component Analysis (PCA) on optimizing machine learning-based IDS using the UNR-IDD dataset.We comprehensively analyzed the performance of Support Vector Machine (SVM), Nave Bayes (NB), and Random Forest (RF) before and after PCA transformation.Experimental results show that PCA significantly reduced the detection latency for SVM and NB without compromising their performance.Specifically, NB + PCA and SVM + PCA achieved a whopping 99.52% and 49.9% reduction in detection latency respectively, making them viable low-latency solutions.However, the PCA transformation did not significantly impact the detection latency of the random forest model.The results demonstrate that NB + PCA is the most efficient and lightweight model for real-time network intrusion detection.These findings demonstrate that PCA is an effective preprocessing step to optimize ML-based IDS for real-time applications.