Performance Analysis of Boosting Algorithms on DDoS Attack Detection with Data Balancing Techniques

Today, a lot of people use the Internet of Things (IoT) in organizations, businesses, and daily life. Attackers have taken advantage of this potential for IoT devices to expose the integrity and security of user data. Especially, the security of the internet and user data is seriously threatened by distributed denial of service (DDoS) attacks. A DDoS attack is a cyberattack, that intends to break down or shut legal traffic from leading to its destination. There are various Machine Learning is employed in this work as classification methods for DDoS attack detection to develop an automated system that can recognize an attack immediately. This research intends a lightweight boosting classification approach to detect DDoS attacks of IoT based on the resampling techniques to minimize the data imbalance. It compares the performance of the proposed approach with AdaBoost, CatBoost, XGBoost, and LightGBM of boosting models based on time, ROC ACU, accuracy, MAE, RMSE, and F1-score. The experimental results demonstrated that the model with 4.079 seconds of the shortest runtime, highest accuracy, and low error rate is XGBoost using the SMOTE technique on the CICDDOS2019 dataset and KDDCUP99 dataset. XGBoost illustrates excellent time efficiency between other models, and it is a pleasant option for scenarios where rapid processing is critical. The accuracy of XGBoost illustrates highest than other models with 0.9998 with Smote and 0.9900 with borderline smote by using CICDDOS2019 dataset as performance. Additionally, it illustrates 0.699 accuracy with smote and 0.9699 of accuracy with borderline smote by using KDDCUP99 dataset. This research proposes an XGBoost classification approach for an effective IoT DDoS attack detection model using SMOTE and borderline SMOTE for balanced resampling.