Comparative Analysis of Anomaly-based Intrusion Detection Model in IoT Network

Abstract

The selection of detection model for Intrusion Detection Sytem (IDS) is crucial as it directly affects the detection performance, especially in IoT networks due to the their constant connection to the internet and their unique properties like heterogeneity. This research aims to study, to develop, and to compare two widely used deep learning anomaly detections models as intrusion detection models, which are: Long Short-Term Memory (LSTM) and autoencoder. IoT-23 dataset is pre-processed, and is used to train and evaluate the anomaly detection models in terms of accuracy, precision, recall, F1 score, and AUC-PR. The results show that LSTM model scored better in most metrics, but the autoencoder model excels at recall. While the findings suggest LSTM is a more balanced approach and autoencoder is a more sensitive approach, future work should explore hybrid approach to combine their strengths to provide more robust detection.