Utilizing the Ensemble Learning and XAI for Performance Improvements in IoT Network Attack Detection

As Internet of Things (IoT) networks continue to expand, it has become increasingly crucial to safeguard the security of these interconnected devices. This research study proposes a novel method for enhancing the effectiveness of IoT network threat detection by employing ensemble learning techniques and Explainable Artificial Intelligence (XAI).

The proposed method involves the utilization of an ensemble model combining an Autoencoder and eXtreme Gradient Boosting (XGBoost), a popular gradient-boosting algorithm. The workflow begins with quantizing the dataset to reduce computational complexity. Subsequently, the autoencoder is trained to learn a compressed representation of the quantized data, while XGBoost simultaneously performs classification tasks. To enhance the efficiency and accuracy of attack detection, feature importance analysis is conducted using XGBoost’s feature importance attribute. This analysis enables the identification of the most influential features, which are then used to prepare a refined dataset, further reducing computational requirements. A Logarithmic layer is introduced within the autoencoder, enabling the linearization of relationships and handling of exponential characteristics.

The novel ensemble model, combining the Autoencoder’s and XGBoost’s strengths, is trained on the refined dataset. This unified model significantly enhances attack detection performance by leveraging the compressed representations learned by the autoencoder and the predictive power of XGBoost. Our proposed model is evaluated in the experiment on the CICIDS2017 data set. The evaluation metrics include accuracy, recall, precision, and runtime. For detection performance, our proposed model achieves an impressive 99.92% detection accuracy on the CICIDS2017 dataset, surpassing most state-of-the-art intrusion detection methods. Moreover, our proposed model exhibits the lowest runtime, further highlighting its efficiency.