Abstract:
The current era has witnessed significant advancements and expansions
in information technology systems. With the daily increase in network communications,
cybersecurity has become a critical issue that demands enhanced threat detection
systems. Researchers have extensively adopted Machine Learning (ML) and Deep
Learning (DL) techniques in the realm of Intrusion Detection Systems (IDS). This
research introduces an innovative method of converting network data into images,
utilizing Convolutional Neural Networks (CNN) for the classification of cyber threat
imagery. It integrates advanced learning methods such as Transfer Learning,
Incremental Learning, and Hyperparameter Optimization (HPO) to streamline
the processing time, facilitate new learning, and sequentially enhance model
performance. The model was evaluated using three standard datasets: NSL-KDD,
UNSW-NB15, and CICIDS2017, achieving an acceptable level of accuracy and precision.
It detects cyber threats comparably to previous research while also incorporating
the ability to learn and detect new cyber threats. This research outlines a prototype
for a cyber threat detection system capable of automatic adaptation for the analysis
and classification of cyber attacks within an environment of continually evolving
cyber threats.