A BINARY AND FAMILY CLASSIFICATION INTRUSION DETECTION FRAMEWORK FOR IOT USING TRANSFORMER CNN HYBRID DEEP LEARNING

Abstract

The proliferation of Internet of Things (IoT) devices has expanded the attack surface for cyber threats, necessitating robust and intelligent Intrusion Detection Systems (IDS). Traditional machine learning models often struggle to capture complex, non-linear spatial and temporal dependencies in high volume network traffic. This research proposes a novel hybrid Deep Learning framework integrating 1D Convolutional Neural Networks (1D-CNN) and Transformer architectures. The 1D-CNN component is utilized to extract local feature correlations from tabular network flows, while the Transformer mechanism captures global context and long-range dependencies. The model will be trained and evaluated on the CICIoT2023 dataset, addressing class imbalance through SMOTE (Synthetic Minority Over-sampling Technique). The proposed framework aims to achieve state-of-the-art accuracy in both Binary Classification (Benign vs. Malicious) and Family Classification (DDoS, DoS, Recon, Web-based, Brute Force, Spoofing, Mirai), providing a granular and actionable security solution for modern IoT environments.