An adaptive machine learning model for IoT security to detect spoofing and DOS attacks

The rising number of Internet of Things (IoT) devices necessitates robust security to counter threats like Denial-of-Service (DoS) and spoofing attacks. Many Intrusion Detection Systems (IDS) leverage Machine Learning (ML), but challenges like concept drift and class imbalance persist. This study presents an adaptive ML-based IDS framework using a Modified XGBoost (Mod-XGB) model. Mod-XGB incorporates a weighted loss function for class imbalance, adaptive instance weighting for concept drift, and a security penalty term to enhance feature selection. Evaluated on the CICIoV2024 dataset, Mod-XGB achieved 97.96% accuracy, outperforming models like Random Forest, Logistic Regression, and AdaBoost. This innovative approach effectively addresses IoT security challenges, offering significant advancements in IDS performance for improved attack detection and resilience in dynamic environments.