An Efficient Feature Selection Technique for Intelligent IDS using Metaheuristics Consensus Ensemble Aggregation.
Authors: S.Vijayalakshmi, Dr.V.Prasanna Venkatesan
DOI: 10.37326/ajsev8.12/2064
Page No: 42-65
Abstract
The mammoth proliferation of digital data across diverse computing platforms, devices and social web portals have dictated the need for efficient pruning and trimming of data to a concise representation. This concise format facilitates the security detection engine to be a winner in any scenario amidst avalanche of impending security threats. This mandates the effective institutionalization of security infrastructure viz. Intrusion Detection System (IDS), Firewalls and Application proxy. Dimensionality Reduction techniques viz. Feature Selection and Feature Extraction mechanism echoes the same sentiment. The inherent bias, variance and weakness exhibited in a single feature selection technique is annulled with the ensemble of feature selection techniques empowered with collective crowd intelligence. This problem effectively translates into optimization problem that recommends the deployment of metaheuristics search algorithm to effectively tackle this NP hard problem with the exponential growth of the problem space and complexity. This proposed model leverages the application of MH algorithms as powerful Ensemble Attribute Aggregator/Fusion agent of the feature selector ensemble outcomes to generate optimal feature combinations that amplifies the classification performance and as well the timing efficiency of IDS. This model encourages the adoption of Ensemble of Filter Feature selectors as it is not tied to any classification algorithm and fast to generate intermediate optimized feature subsets that is fed to the IDS model. This model promises to deliver quick real time decisions with high TPR and minimal detection latency. Using MH as aggregators helps to avoid myopic decisions engineered by Filter Ensemble. The data deluge property in Ultra High Dimensional Dataset (UHDD) is effectively countered with MH algorithms that optimizes the feature aggregation to address the scalability and stability issues with increased execution speed up and convergence rate. Various evaluation metrics have been proposed to comprehend the efficacy of this model with NSL-KDD dataset comprising different attacks.
