Efficient Data Preprocessing for Deep Learning-Based Intrusion Detection Systems Using the CICIoT2023 Dataset

Abstract

The proliferation of Internet of Things (IoT) devices has introduced significant cybersecurity challenges, necessitating robust intrusion detection systems (IDS). This paper presents an efficient data preprocessing framework tailored for deep learning-based IDS using the CICIoT2023 dataset. Our approach integrates advanced preprocessing techniques including intelligent feature selection, multi-stage normalization, dimensionality reduction via principal component analysis (PCA), and adaptive synthetic minority oversampling (ADASYN) to address class imbalance. We evaluate three deep learning architectures: Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) networks. Experimental results demonstrate that our preprocessing pipeline significantly enhances detection performance, achieving 98.74% accuracy, 98.32% precision, 97.89% recall, and 98.10% F1-score with the DNN model, outperforming baseline approaches by 12.3% in accuracy and reducing false alarm rates by 41.2%. The proposed framework exhibits exceptional efficiency in handling large-scale, imbalanced IoT traffic data while maintaining computational feasibility.