INTRUSION DETECTION USING DEEP (CNN) CONVOLUTIONAL NEURAL NETWORK FEATURE EXTRACTION WITH (EPCA) ENHANCED PRINCIPAL COMPONENT ANALYSIS FOR DIMENSIONALITY REDUCTION

Abstract

(IDS)Intrusion detection system, is an extremely significant to prevent network attacks, and in classification of network traffic to determine anomalies inside network. However, no any existing studies, explored an efficient IDS, to address the problem of low accuracy engendered by redundant, irrelevant, non-linear features and dealing with large dataset. Hence to overcome this issue, different level of features traversed in several hidden layers undergoes deep learning and extracted by two 1-dimensional models of Deep CNN. Then with the integration of (Fourier)F-transform and (PCA)Principal component-analysis transforms the non-linear features to linear feature sets, and reduces high dimensionality to low-dimensional features. Hence it aids in minimising the PCA computation time and increases model robustness. This sort of enhanced PCA with F-transform facilitates to increase accuracy of classification. The transformed features are classified efficiently through algorithms (RF)Random Forest, (GNB)Gaussian Naive Bayes, XGBoost and KNN classifier. The comparative assessment of proposed IDS model, outperformed in classifying normal and abnormal data with higher accuracy.