Adaptive intrusion detection in IoT networks: Leveraging honeypots and synthetic data for attack mitigation
Rupal Chaturvedi, Neha Tyagi, Nikhil Panwar, Nitin Verma · 2025
Internet of Things (IoT) devices are attracting more and more cyber threats, resulting in them being vulnerable to intrusion assaults as they are swiftly expanding. This research focuses on enhancing IoT network security by leveraging honeypot-based intrusion detection systems improved with synthetic data generation. The system detects illegitimate individuals through a machine learning based anomaly detection methodology. To deceit the attacker and avoid illegal access to real data, GAN is used to generate realistic synthetic sensor data to mislead the attacker. We have developed an intrusion detection system using Feed Forward Neural Network which is trained with UNSW-NB15 dataset and achieved a remarkable result with 99.56%, 90.53%, 96.40%, and 93.38% as Accuracy, Precision, Recall, and F1-score, respectively. Fully connected DNN-based GAN, which is trained on IoT_Weather dataset from ToN_IoT, generated artificial data and is evaluated and achieved Mean Absolute Error as 14.8207, KS Test Stat as 0.2326, JSD as 0.4695, Cosine Similarity as 0.9151, Feature Correlation Difference as 0.2551. Integrating FFNN based intrusion detection and GAN-based synthetic data generation presents a flexible and adaptable intrusion detection architecture guaranteeing enhanced resistance to changing cyber threats.