Comparison of Ensemble Learning Methods on the IoT-23 Dataset

The Internet of Things (IoT) has provided numerous benefits across various sectors, but it also poses significant challenges in cybersecurity, particularly malware threats. Malware on IoT devices has the potential to damage systems, steal data, and disrupt network performance. Previous research has shown that the Na ̈ıve Bayes algorithm produces a low accuracy of 0.24, increasing slightly to 0.35 when combined with AdaBoost, and reaching 0.99 when combined with XGBoost using the soft voting method. However, there is still room to explore other ensemble learning methods to obtain more stable results. This research focuses on the application of an alternative ensemble learning method, namely stacking, using the IoT-23 dataset with reference to the CRISP-DM framework. The results show that the stacking method can significantly improve malware detection accuracy from 0.35 to 0.72, thus proving superior to soft voting and can be an effective approach in improving malware detection performance in IoT networks.