Improving the Performance of Xgboost in Multi-Class Iot Attack Detection With Loss Function Optimization

The rapid growth of Internet of Things (IoT) makes the networks to which IoT devices are connected increasingly vulnerable to cyberattacks. This threat increases the importance of robust intrusion detection systems for IoT networks. We propose a novel XGBoost-based classification model with an optimized loss function. Furthermore, a fitness function was designed to minimize the multi-class loss function mlogloss and applied hyperparameter optimization with Optuna in task. To evaluate the effectiveness of the proposed model, we used the comprehensive CICIoT2023 dataset to effectively identify a wide variety of attacks in IoT environments. The proposed model configured with optimal hyperparameters achieved remarkable classification results with 99.39% precision, 99.41% recall, 99.39% F1-score, and 99.33% accuracy. The results demonstrate that the proposed model is a robust and feasible alternative for analyzing complex IoT network traffic. Optimizing the loss function in multi-class scenarios allows the model to produce more balanced and reliable predictions. Additionally, integrating XGBoost with Optuna provides an effective strategy for IoT security applications. Consequently, the proposed approach not only effectively distinguishes between various types of attacks in IoT traffic but also highlights the potential benefits of task-specific loss function optimization strategies for future cybersecurity models.