An Adversarial-Resilient Intrusion Detection System for IoT Networks Using Hybrid Learning and Causal Analysis

With the rapid expansion of the Internet of Things (IoT) in critical infrastructures such as smart cities, healthcare, and transportation, securing IoT networks has become a pressing challenge. Due to the distributed and resource-constrained nature to adversarial attacks that manipulate input data through subtle perturbations, misleading machine-learning models. This study proposes a novel adversarial-resilient intrusion detection system (IDS) framework that combines Stacking Ensemble Learning with Causal Consistency Checking. The system integrates multiple deep models, Feedforward Neural Network (FFNN), Long Short-Term Analysis. Memory (LSTM), Convolutional Neural Network (CNN), and Autoencoder, whose outputs are aggregated through an XGBoost meta-learner along with causal consistency scores. Evaluations on the Bot-IoT dataset under various adversarial scenarios (IoTGAN, HAA, and rule-based attacks) demonstrate superior detection accuracy and robustness. The proposed design effectively mitigates black-box adversarial attacks, ensuring high reliability for IoT security.

Keywords:

Intrusion Detection , Internet of Things , Adversarial Attacks , Ensemble Learning , Causal Analysis , Network Traffic