Hybrid Autoencoder and Isolation Forest for IoT Anomaly Detection with a Novel Model

Securing the Internet of Things (IoT) in everyday life remains a significant challenge, which makes anomaly detection in these devices both important and necessary. In this work, we propose a hybrid approach that employs Autoencoders (AEs) for feature extraction and Isolation Forest (IF) for anomaly identification. To address errors caused by variations in sensor data, we further introduce a whitening method, which normalizes input features before detection. Experiments on the CIC IOT-DIAD 2024 dataset show that the hybrid AE+IF method achieves an accuracy of 0.98, outperforming either technique used independently. Incorporating covariance information through the whitening method further improves performance to an accuracy of 0.99. We compared the results with other datasets, such as N-BaIoT, TON_IoT, and UNSW-NB15, and observed that the model also delivers good performance on these datasets. Overall, the study provides a practical and interpretable framework that can be scaled for deployment in real IoT environments.