Performance optimization for Intrusion Detection by Long Short Term Memory (LSTM)

Abstract

Concerns about cyber threats have emerged as the expansion of system connectivity and the proliferation of system applications intensified in the industry. This has underscored the necessity for a robust defense mechanism against various cyber threats, including potential intrusions from malicious actors within the network. A specially targeted system is the intrusion detection system (IDS), designed to safeguard the confidentiality, integrity, and availability of network traffic, especially in critical sectors like healthcare. Recent advancements in the area of IDS involve the utilization of artificial intelligence (AI) and deep learning (DL) based IDS to efficiently recognize network issues. Notably, the research at hand adopts a deep learning approach employing Long Short Term Memory (LSTM) models, applied to the CICIDS-2019 dataset that is sourced from New Brunswick University’s website. The focal point of evaluation lies in the precision, recall, F1-score, and accuracy metrics, specifically analyzing its performance in identifying Denial-of-Service (DoS) cyber-attacks. The findings of this study lighten the superior performance of the Long Short Term Memory method in the realm of intrusion detection systems. The LSTM model showcases its proficiency, particularly in discerning Denial of Service attacks by giving a loss of less than 0.03%.