A comprehensive survey on intrusion detection algorithms

Although there are many reviews on Intrusion Detection Systems (IDS), the basic parts of Intrusion Detection Algorithms (IDA), such as imbalanced datasets, feature engineering, and model design, have not been fully studied. This review thoroughly examines modern IDA, emphasizing recent progress, current challenges, and potential future research paths. First, we explore three different strategies to handle imbalanced datasets: resampling, Synthetic Minority Over-sampling Technique (SMOTE), Generative Adversarial Networks (GAN). Next, we examine a few key feature extraction techniques, including Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Autoencoder (AE), among others. Additionally, we explore filtering, wrapper, and embedded methods for feature selection. Then, we explore model design approaches for IDA, considering both ensemble and non-ensemble learning methods. We provide a thorough assessment of ensemble techniques: bagging, boosting, and stacking. We also evaluate a variety of non-ensemble methods, including Naive Bayes (NB), K-Nearest Neighbors (KNN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), among others. Finally, we briefly outline relevant applications, challenges and future research directions. This survey will serve as a valuable resource for researchers and practitioners, and foster the advancement of IDA technology.