Hybrid Deep Learning-Based Autoencoder-DNN Model for Intelligent Intrusion Detection System in IoT Networks

The internet of things (IoT) has revolutionized contemporary life by seamlessly integrating devices and applications across various fields, including precision agriculture, digital healthcare, intelligent transportation, and smart urban ecosystems. Nevertheless, IoT infrastructures are highly vulnerable to sophisticated cyber threats, demanding the implementation of resilient and advanced security mechanisms. Advanced deep learning methodologies have become a transformative force in intrusion detection, effectively overcoming the constraints of traditional signature-based techniques, especially in identifying previously unseen zero-day threats. To better security violation recognition in IoT surroundings, this research introduces an innovative integrated deep learning method that unites an Autoencoder and a deep neural network (DNN). Through the application of adaptive synthetic sampling (ADASYN) approach, edited nearest neighbor (ENN), and intentionally allocated class weights, the presented system efficiently alleviates uneven class representation, a natural obstacle to exact irregularity recognition. Empirical assessment on the NF-ToN-IoT-v2 data collection indicates that the introduced strategy secures a 98.08% (accuracy) in dual-class classification, illustrating its outstanding potential for anomaly recognition and exceeding current approaches. This advancement establishes a resilient safeguard for IoT infrastructures against continuously evolving cyber threats.