Interpretable And Proactive Intrusion Detection Using Discrete Optimization Learning: Futuristic Approach
Main Article Content
Abstract
Early detection of Network security relies heavily on the detection of intrusions, yet existing methods often struggle to identify threats before a session concludes. This limitation stems from the predominant use of features extracted from entire sessions, hindering early detection. AI based interruption location frameworks have arisen as an essential device in this space, although the challenge of designing an optimal framework persists. To address this issue, a novel approach is proposed, leveraging packet data as features to discern malicious traffic. However, this method introduces the risk of false positives, where normal packets may be erroneously classified as intrusions, and vice versa. To counteract this, the proposed method focuses on learning patterns of packets that are uninformative for distinguishing between intrusions and benign sessions. Through extensive experimentation, it has been demonstrated that this approach enables early detection of intrusions, even before session termination, while maintaining detection performance comparable to established methods. This innovative strategy represents a significant advancement in enhancing network security. In this we are using CICEV2023 Ddos Attack data set it also provide us an distributed threats that which we can easily remove from the original dataset and considered as a cyber threat by using Discrete optimization learning based on the LSTM (Long short Term Memory) and Back propagation techniques for retrieve the process if any miscalculation occurs. With these techniques, we acquire the accuracy rate of 96.14% and 84.6% recall as well the main achievement of this project is to detect the intrusion before the session gets terminated. The NIDS is crucial for network security, especially when utilizing ML and DL technologies to combat complex attacks. Our article presents another two-stage interruption identification framework involving circulated profound learning for ongoing investigation. This system excels in detecting distributed malicious activities and employs a hybrid model for precise attack identification. Additionally, our model has broad applications in various DL fields and demonstrates improved training loss rates through effective data cleaning techniques.