Effective Combining of Feature Selection Techniques for Machine Learning-Enabled Intrusion Detection System

This paper presents a methodology combining various feature selection methods for enhancing the accuracy of machine learning-based intrusion detection, focusing on the NSL-KDD dataset. The NSL-KDD dataset is a widely used benchmark for training and testing intrusion detection models [2]. It includes various subsets (e.g., KDDTrain+, KDDTest+), providing a balanced mix of records to support different machine learning techniques. Our research seeks to identify the most relevant features in the dataset for accurately pinpointing areas of network intrusion or attack. The process starts with effective data preprocessing, utilizing different formats in the NSL-KDD dataset, which includes binary labels, attack-type labels, and difficulty level annotations (in ARFF and CSV). This step is crucial for ensuring clean data for feature selection and model training. By removing redundant features, we aim to improve classifier performance and efficiency. Central to our strategy is enhancing attack detection through the Hippopotamus algorithm [6], an advanced feature selection technique that identifies the most critical features for specific attack categories. Our comprehensive feature selection strategy can be extended to other datasets and cybersecurity applications, paving the way for more sophisticated and reliable intrusion detection mechanisms in modern networks.