An Adaptive Intrusion Detection System for Evolving IoT Threats An Autoencoder FNN Fusion

The rapid expansion of the Internet of Things (IoT) has introduced a vast attack surface, resulting in diverse and evolving cyber threats that challenge conventional Intrusion Detection Systems (IDS). This project, titled “An Adaptive Intrusion Detection System for Evolving IoT Threats using Autoencoder–FNN Fusion,” presents a hybrid deep learning approach that enhances detection accuracy and adaptability against dynamic IoT attacks. The proposed framework integrates an Autoencoder (AE) for feature compression and noise reduction with a Feedforward Neural Network (FNN) classifier for intelligent threat categorization. The Auto encoder learns deep latent representations of IoT traffic data, capturing intrinsic patterns that help detect both known and zero-day attacks, while the FNN refines classification with improved generalization. A Random Forest model is also employed as a baseline to validate performance consistency and robustness. The system achieves adaptive learning by periodically retraining with new traffic data, ensuring resilience to evolving threat behaviors. Extensive experimentation demonstrates that the AE–FNN model achieves high detection accuracy and reduced false alarm rates compared to traditional machine learning models. The developed web interface allows users to upload IoT datasets, visualize predictions, and analyze detected attack types effectively. Overall, this adaptive AE–FNN Fusion IDS offers a scalable, intelligent, and future-ready solution for safeguarding IoT environments against emerging network intrusions.

Keywords: Intrusion Detection System, IoT Security, Autoencoder, Feedforward Neural Network, Deep Learning, Random Forest, Adaptive Learning, Cyber Threat Detection.