Hybrid Models for IoT Security: Tackling Class Imbalance
Yassine El Yamani, Yousra Fadili, Jihad Kilani, Najib El Kamoun, Youssef Baddi, Faycal Bensalah · 2024
This study introduces a new method to enhance IoT security, focusing on the “SimpleHome XCS7 1002 WHT Security Camera”. We use a CNN-LSTM hybrid model combined with balanced resampling techniques to tackle class imbalance in security datasets. Our testing on the “N-BaIoT” dataset, which includes data from IoT devices targeted by Mirai and BASHLITE botnet attacks, shows that our approach improves precision from 92.00% to 93.08% and recall from 88.00% to 90.79%. The F1-score increases to 87.92 %, accuracy to 90.70%, and the loss decreases to 0.1372. These results demonstrate how our method can significantly enhance the detection of attacks in IoT systems. This research shows the effectiveness of combining deep learning with resampling strategies for IoT security, supporting the development of adaptable and resilient security solutions for our increasingly connected world.