Generating Network Intrusion Image through IGTD Algorithm for CNN Classification

Stealthy attacks make Intrusion Detection System (IDS) research perennial and notably important. From the literature, a few problems were identified in IDS studies particularly with regards to image generation. Image generation is the preprocessing phase during the development of CNN classifier. Simple feature extraction process that are applied in a few studies may lose features’ details. They disregard spatial information such as pair pixels information. Secondly several public datasets suffer from realistic network traits. This study will apply the improvised version of IGTD algorithm to resolve image generation process. IGTD algorithm has been adopted in Cancer Cell Lines (CCLs) study. Realistic dataset of CIRA-CIC-DoHBrw-2020 and CICIDS2017 will be run over the algorithm. These dataset have been reviewed to be the closest realistic production network dataset. Result shows the generated image has notable difference between attack and benign traffic. The highest trained classification model has achieved almost 80% of accuracy rate.