An Adaptive Intrusion Detection System for Evolving IoT Threats: An Autoencoder-FNN Fusion
J. Jasmine Shirley, M. Priya · IEEE Access · 2025
The increasing number of sophisticated attacks targeting the Internet of Things environment highlights the critical importance of a strong intrusion detection system. This study proposes a specialized IDS classification approach tailored for IoT networks to tackle their evolving and diverse threat landscape. The proposed model combines an Autoencoder for feature extraction and a Feedforward Neural Network for classification. This combination enhances the precision and efficiency of intrusion detection. The Autoencoder effectively captures relevant features from the complex and high-dimensional IoT network traffic data, improving the subsequent classification process by reducing dimensionality. On the other hand, the Feedforward Neural Network excels in distinguishing between normal network behavior and various intrusion patterns. Its ability to handle nonlinear relationships in data is leveraged for this purpose. This fusion of techniques provides greater adaptability to evolving threats and dynamic network conditions, enabling the IDS model to identify instances of attack while minimizing false positives accurately. The rigorous evaluation demonstrates the robust performance of the model, achieving an impressive accuracy of 99.55% in binary class classification and 90.91% in multiclass classification. Metrics such as precision, recall, and F1-score confirm the model’s capability to detect diverse intrusion patterns. High ROC AUC values also highlight its effectiveness in handling various attack scenarios. This research significantly contributes to IoT network security by providing a reliable and efficient IDS model designed to tackle interconnected environments’ unique challenges. It positions itself as a valuable tool for real-world intrusion detection scenarios.