FLOW PATTERN CONSTRUCTION-BASED DECENTRALIZED INTRUSION DETECTION FRAMEWORK USING TL-IML3CSTM AND GAUSSIAN-Π FUZZY

Abstract

A Decentralized Intrusion Detection System (DIDS) distributes detection tasks across multiple nodes and collaboratively identifies the network intrusions using data patterns. However, the traditional techniques failed to map the data flow patterns after the identification of normal and generic attacks, thus causing misclassifications or inaccuracies in the system. Hence, to overcome these issues, the HamKowski Ford-Fulkerson Algorithm (HKFFA) and Transfer Learning-based Inverse Mish Le-LayerCun Long Short-Term Memory (TL-IML3CSTM) are used in this framework. At first, the data is collected, preprocessed, and then balanced to reduce the training duration. After that, the behavioral patterns are identified using EGDBSCAN and flow patterns are constructed utilizing HKFFA to enhance the attack detection process. The normal and generic attacks are then categorized with high Attack Detection Accuracy (ADA) of 99.32% using TL-IML3CSTM. After that, the low-severe attacks and normal data are mapped with possible flow patterns. Here, the mapped data is preserved and stored in the cloud for future usage, while the non-mapped data is blocked for security purposes, thus outperforming the prevailing models.