Network Intrusion Detection Using Semi-Supervised Learning for Unlabeled Big Data Through Oversampling

Network intrusion detection systems help identify malicious actors and activities within the network, especially in interconnected networks. To detect abnormalities, potential cyber threats, and malicious activities, the Machine Learning (ML)-based approaches have considerable potential. As the amount of data grows, a dimension reduction technique is required, while training the ML models. Additionally, a large portion of the available data remain unlabeled most of the time. Taking all these into consideration, we introduce a semi-supervised learning-based network intrusion detection technique that can deal with partially-labeled data. More specifically, we utilize Random Oversampling (RO) to rectify data disparity and Principal Component Analysis (PCA) to reduce data dimension. Three semi-supervised learning algorithms - Label Spreading, Self Training, and Transductive Support Vector Machine (TSVM),…