Improving Generalization of ML-Based IDS With Lifecycle-Based Dataset, Auto-LearningFeatures, and Deep Learning

The increasing complexity and volume of network attacks necessitate robust intrusion detection systems (IDS). Current machine learning (ML)-based IDS often suffer from limited generalization capabilities when faced with dynamic attack behaviors. This research proposes an enhanced ML-based IDS framework utilizing lifecycle-based datasets, auto-learning features, and deep learning techniques to improve the detection of novel intrusions. The existing system leverages convolutional neural networks (CNN), but we explore alternatives such as decision trees, random forests, multilayer perceptrons (MLP), and XGBoost to enhance performance. By incorporating lifecycle datasets and automatic feature learning, this approach aims to improve the adaptability, accuracy, and scalability of IDS. Empirical evaluations reveal that the proposed system significantly enhances generalization and detection rates, providing a more resilient defense against sophisticated cyber threats.

Keywords: Intrusion Detection System (IDS), Lifecycle-Based Datasets, Decision Tree, Random Forest, XGBoost, MLP.