Small-Scale Dataset for Machine Learning Based Network Intrusion Detection

An intrusion detection system (IDS) was introduced to detect various intrusive attempts on the network. Several machine learning (ML) techniques have been incorporated to build robust and intelligent IDS. Researchers are maximizing public intrusion datasets, such as the KDD CUP 99, NSL-KDD and the CICIDS-2018 datasets, to train the ML algorithms. However, public intrusion datasets have large data sizes, which cause a high training time for the ML algorithms. Moreover, public intrusion datasets have high-dimensional input features, thus requiring tedious feature extraction methods. Furthermore, these datasets have an imbalance ratio between normal and attack traffic, which leads to less efficient training supplied to the ML algorithms, resulting in low accuracy. An intrusion dataset that can reduce the training and processing time whilst having a balanced dataset between normal and attack traffic must be proposed to overcome these issues. Hence, a small-scale and robust intrusion dataset is developed in this paper to train the ML classifier. Simulation results show that the proposed dataset improves the accuracy of the ML classifier by 19.15%, 99.26% and 40.48% considering accuracy, recall and F-measure, respectively, compared with the CICIDS-2018 public dataset.