Comprehensive Analysis of Different Boosting Techniques for Attack Detection in IoT Network

IoT is a fast-growing field with applications in all sectors, but a threat to, availability, confidentiality, and integrity of data is the main hurdle to the growth of IoT. There are two main types of IoT attack detection related intrusion detection systems, specifically signature-based and anomaly-based IDSs. CIE-CICIDS2018 dataset is used for experimentation with attack detection. It is an imbalanced dataset with maximum benign traffic and very little attack traffic. Ensemble learning is the most powerful technique for handling imbalanced datasets with different bagging and boosting techniques. In this research work four boosting techniques, Adaboost, GBM, light GBM, and XGBoost are applied to CIE-CIC-IDS2018 and Performance at various learning rates—0.05, 0.075, 0.1, 0.25, 0.5, 0.75, and 1—is compared. Once the optimal learning rate based on accuracy for the dataset has been established, performance measures, mainly F1 score, precision, recall, and accuracy, of all four boosting methods are compared with the optimal learning rate. It has been found that XGBoost is giving the best results among all, with the highest learning rate of 0.25. The suggested boosting method’s outcomes are contrasted with those of cutting-edge systems that employ Random Forest, SVM, LSTM, and KNN. It has been observed that all four boosting techniques are giving better results as compared to existing ML and DL models.