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

The increasing sophistication of attacks targeting the Internet of Things environment underscores the critical need for robust intrusion detection systems. This study introduces a specialized IDS classification approach to address IoT networks’ evolving and diverse threat landscape. The proposed model integrates an Autoencoder for feature extraction with a Feedforward Neural Network for classification, creating a fusion that enhances the precision and efficiency of intrusion detection. The Autoencoder captures and reduces the dimensionality of high-dimensional IoT network traffic data, and the FNN distinguishes between normal network behaviour and various intrusion patterns by leveraging its ability to model nonlinear relationships. This dual-stage architecture offers adaptability to dynamic network conditions, accurately identifying attack instances while minimizing false positives. Rigorous evaluations demonstrate the model’s robust performance, achieving a remarkable accuracy of 99.55% in binary classification and 90.91% in multiclass classification. Key metrics such as precision, recall, F1-score, and high ROC-AUC values validate its effectiveness in detecting diverse intrusion patterns and handling various attack scenarios. Additionally, the study employs a two-stage data balancing strategy that significantly enhances the detection of minority class attacks. By addressing critical challenges in IoT network security, including imbalanced datasets and evolving threats, this research contributes a reliable and efficient IDS model optimized for real-world applications. The AE-FNN model’s high accuracy, robustness, and scalability position it as a valuable tool in securing interconnected IoT ecosystems, making it an important advancement in network security.