A Hybrid Deep Learning Model for IoT Network Anomaly Detection
Yonas Getachew Mulissa, Wei Li, Ajoy Kumar, Ling Wang · 2025
The Internet of Things (IoT) has transformed digital device communication, creating a critical need for effective anomaly detection (AD) systems to ensure secure network operations. Deep Learning (DL) has proven to be a highly effective solution for AD, outperforming traditional machine learning approaches due to advancements in neural network designs, enhanced hardware capabilities like GPUs, and the availability of large training datasets. This study introduces a hybrid deep learning model that leverages the Long Short-Term Memory (LSTM) network's ability to capture short-term temporal patterns and the Transformer's attention mechanism for contextual understanding and efficient parallel processing. The proposed hybrid LSTM-Transformer model aims to improve classification accuracy, reduce computational demands, and optimize key metrics such as accuracy, loss, precision, False Positive Rate (FPR), and F1-Score for multi-class scenarios. Experimental results of comparing the hybrid learning models with other hybrid models formed by combinations of CNNs, LSTM, and Transformer architectures have shown promising results, highlighting the potential of the proposed LSTM-Transformer model to enhance IoT anomaly detection by capitalizing on the strengths of both LSTM and Transformer architecture.