Optimized Deep Learning Mechanism for Intrusion Detection: Leveraging RFE-Based Feature Selection and PCA for Improved Accuracy

With the growing demand for wireless networks, network traffic has surged, particularly involving large-scale data transmission. This results in an increased number of network security threats and the emergence of various cyber-attacks. To overcome these threats, the development of an efficient intrusion detection system (IDS) has become crucial. By leveraging the recent advancements in Artificial intelligence (AI), this project introduces an optimized deep learning mechanism utilizing a prevalent deep learning model, the Deep Neural Network (DNN) algorithm. This proposed method uses feature engineering techniques like feature selection and feature reduction. Given the high dimensionality of the selected datasets, these techniques are employed to prevent over-fitting. Recursive feature elimination performs feature selection followed by Principal Component Analysis for the feature reduction process, thus enhancing the model's efficiency. The system is assessed using two benchmark datasets: KDD and UNSW-NB15. The experimental results indicate that the presented DNN model on the KDD dataset has achieved 92.45% accuracy and 91.35% accuracy on the UNSW-NB15 dataset for intrusion classification.