Enhancing Intrusion Detection Systems through Advanced Feature Engineering and Machine Learning Techniques

This study investigates enhancing intrusion detection systems (IDS) through machine learning and advanced feature engineering. I evaluate classification algorithms, including k-nearest neighbors (KNN) and support vector machines (SVM), on the ADFA-LD 12 dataset, finding that SVMs, especially with two-sequence feature spaces, outperform other methods. A detailed analysis of linear and sigmoid kernels shows that the two-sequence space improves detection accuracy, with recursive feature elimination identifying optimal performance with fewer than 240 features. One-class SVMs, used for outlier detection, perform poorly. The study underscores the potential of SVMs and suggests further research into scalable algorithms and improved kernel methods for unsupervised detection.