Deep Learning-Based Network Intrusion Detection for IoT Using CNN and Transformer Models

The unprecedented proliferation of the Internet of Things (IoT) has transformed modern communication systems but introduced profound cybersecurity challenges due to the heterogeneous and resource-constrained nature of IoT devices. While deep learning offers promising solutions, achieving a balance between high detection accuracy, computational efficiency, and resilience to novel threats remains an open problem. This study advances the state-of-the-art in hybrid deep learning for IoT intrusion detection by proposing a parallel CNN-Transformer architecture that emphasizes sustainability, zero-day threat detection, and real-world deployability. The model leverages the complementary strengths of CNNs for local spatial feature extraction and Transformers for global temporal dependency modeling. Evaluated on the CICIDS2017, Bot-IoT, and the contemporary CIC-IoT-2023 datasets, the proposed model achieves an average accuracy of 98.7%, 99.1%, and 97.8%, respectively, outperforming baseline models including Random Forest, CNN-only, Transformer-only, and recent hybrids like CNN-LSTM. Crucially, it demonstrates robust zero-day detection (86.0% average detection rate in leave-one-class-out experiments) and maintains a low false positive rate (