A Comparative Analysis of CNN, LSTM, and Autoencoder Models of IoT Intrusion Detection
Shwe Sin Myat Than, Akari Myint Soe, Aung Htein Maw · 2024
With the rise of Internet of Things (IoT) networks, the need for faster, complex and optimized anomaly detection system to protect the network is more important. This research paper offers a comparison between three emerging deep learning algorithms: Autoencoders, LSTM networks, and CNN towards anomaly detection in IoT setting. Using dataset, IoT-23, we demonstrate these techniques using several evaluation metrics such as accuracy, Mean Absolute Error (MAE) and Precision-Recall Area Under Curve (PR-AUC). CNNs were the top-performing model with the highest PR-AUC, the lowest MAE, and significantly higher accuracy which we set as the new record in the field. When tested and evaluated, LSTMs had better accuracy and PR-AUC as compared to other algorithms such as Autoencoders which were also found to be useful.