ROBUST DEEP LEARNING MODELS FOR SECURE NETWORK TRAFFIC ANALYSIS FOR IMPROVE CYBERSECURITY

Abstract

The fast growth of internet-connected devices has made the cyber landscape more complicated and presents new challenges for security experts seeking intelligent and real-time solutions to combat these threats. A novel architecture for deep learning (DL) is proposed by this framework, which employs ML methods and analysis of network data to thwart cyber-attacks. This study presents a CNN and LSTM model-based DL-based network intrusion detection system (NIDS) that effectively uses the CICDDoS2019 dataset to address this issue. Before training a model efficiently, data must be preprocessed by removing unnecessary or redundant characteristics, normalizing it, encoding labels, and dividing it into a training and testing set. When it comes to accurate intrusion detection, CNN models are employed for automated spatial feature extraction, whereas LSTM models are used for modeling temporal and sequential traffic patterns. Accuracy (ACC) levels of 99.96% for a CNN model and 99.98% for an LSTM model were achieved in the experimental findings, which also demonstrated good Precision (PRE), Recall (REC), and F1 score (F1) values. Training and validation performance curves, as well as confusion matrix analysis, demonstrated that the proposed models are reliable and robust, with few classification errors and minimal overfitting. Additionally, the suggested framework is contrasted with alternative DL and ML techniques, which further prove its excellence in analyzing network data and detecting cyberattacks. The proposed system would enable effective and reliable cybersecurity monitoring in contemporary network environment.