Securing Bluetooth Technology Against BrakTooth Vulnerabilities Through Hybrid Machine Learning Model
Jahanggir Hossain Setu, Md. Shazzad Hossain, Nabarun Halder, Ashraful Islam, M. Ashraful Amin · 2024
BrakTooth, which impacts various versions (including 5.2) of Bluetooth technology, represents a substantial security threat across multiple sectors, including IoT, automotive, healthcare, and financial systems. This study utilizes a hybrid Machine Learning (ML) classifier, combining Extra Trees, Adaptive Boosting (AdaBoost), and Random Forest (RF), to analyze the ISOT BrakTooth Attack dataset which focuses on the baseband layer and Link Manager Protocol (LMP) in the Bluetooth protocol stack. The study proposes a methodological framework that integrates advanced data preprocessing and resampling techniques including the Combined Cleaning and Resampling (CCR) method, Synthetic Minority Over-sampling Technique (SMOTE), and Gaussian SMOTE to address class imbalance issue. Information Gain was employed as a feature selection technique to find the most contributing features in terms of predicting outcomes. The performance of the classifier was evaluated across different distributions of the dataset, with the CCR technique showing remarkable improvement in all metrics. It achieves an accuracy of 99.44%, and precision, recall, and F1-score values above 99.78%. This result demonstrates a substantial enhancement over the original dataset distribution which achieved an accuracy of 90.97% without resampling or feature selection. Moreover, the Area Under Curve (AUC) score of 0.98 confirms the effectiveness of the proposed methodology. The results highlight the importance of resampling techniques in enhancing the detection capabilities of ML classifiers against Bluetooth security threats.