Detection of Assaults in Network Intrusion System using Rough Set and Convolutional Neural Network

The exponential growth of network data traffic has posed significant challenges to conventional intrusion recognition systems. These systems often struggle with feature selection and classification algorithms, leading to inefficiencies in processing large datasets. Additionally, biased raw data traffic can negatively impact classification outcomes. To address these challenges, we propose a novel network intrusion recognition model that combines rough set theory with the You Only Look Once convolutional neural network (YCNN). Our main purpose is to enhance the accuracy and efficiency of intrusion detection in large networks. Our primary contributions include: (1) transforming tabular data into an image format, allowing YCNN to process data more effectively by placing similar attribute features in adjacent pixel locations; (2) employing rough set theory for feature selection to reduce the dataset size and improve YCNN accuracy; (3) addressing data bias by assigning cost function weight coefficients based on category frequency, thereby reducing the false alarm rate; and (4) eliminating the need for feature engineering, thus saving time and resources. We evaluated our model using the CSE-CIC-IDS 2018 dataset, achieving superior accuracy and lower computational cost compared to traditional models. These results demonstrate that our approach is both efficient and reliable for intrusion detection in large networks. The global implications of our research include providing a scalable and accurate solution for network security, applicable in various fields such as natural language processing, object detection, and image recognition.