Toward a new IDS based on PV-DM (Paragraph Vector-Distributed Memory Approach)

Abstract Intrusion detection systems (IDS) have become increasingly important in recent years as a means of assuring network security and eliminating undesirable conduct. Intrusion detection systems (IDSs) are either software applications or hardware devices that monitor network traffic for indications of malicious activity or policy violations. The study offers a detection technique based on feature selection, the Distributed Memory Paragraph Vector (PV-DM), and machine learning clas-sifiers. We evaluated our proposed system using the NSL-KDD and UNSW-NB15 datasets. Experiments with multiclass classification demonstrate the efficacy of our approach, which achieves 98.92 percent accuracy, 98.92 percent precision, 95.44 percent recall, and 96.77 percent F1-score with only four features in NSL-KDD and 82.86 percent accuracy, 84.07 percent precision, 77.70 percent recall, and 80.20 percent F1-score with only six features in UNSW-NB15. Additionally, our proposed approach outperforms the standard LSTM model across the board.