Semi-Supervised Multi-Instance Neurologic Adaptive Learning Intrusion Detection System

Abstract

In the earlier work, Single Instance Single-Label Learning mechanisms are proved faster and converge for better results but the accuracy is the concern in the Intrusion Detection systems (IDS) in large scale data centric networks. Evolution of supervised training mechanisms have favored more accurate decision making, however failed to incorporate the dynamism of new instances which are in early stages and non-overlapping with current training data sets. This paper proposes a cascaded and hybridized mechanism involving Semi-supervised Multi-Instance Neurologic Adaptive Learning (SMI-NAL) mechanism. The objective is to estimate the threshold level of convergence and improve the system accuracy. In addition, the proposed learning mechanism reduces the computational complexity to achieve proactive measure which is the real challenging phenomenon. Compared to the conventional IDS process, the proposed learning mechanism improves the accuracy of IDS and updates the machine learning set (Training set) more appropriately and wrong decision making can beeliminated. The key features of IDS such as robustness, scalability and ease of use are achieved using the proposed learning mechanism.