Abstract
Internet of Things (IoT) technology and its applications have new cyberattacks requiring powerful and effective cybersecurity solutions. Although machine learning (ML) techniques are promising to defend cyberspace, most of them depend on predefined hyperparameters, which restrict their effectiveness in defending against a dynamic and evolving threat landscape. To tackle this, we introduce a learning-based framework, subsequently, the integration of an improved hyperparameter optimization technique enhances cybersecurity in IoT domains. Specifically, it exploits an Enhanced Bayesian Optimization (EBO) approach for the optimization of ML models used in attack detection. This technique captures adequate features like tuning covariance hyperparameter dynamically, acquisition functions, parallelization, and cost modeling. In this paper, we propose an algorithm called Learning-based Method with Hyperparameter Optimization for Cyber Attack Detection (LbMHO-CAD), which combines EBO and different ML models, such as Decision Tree, K-Nearest neighbors, Logistic Regression, Support Vector Machine, and Random Forest. These models were evaluated for their generalization ability in detecting a range of cyberattacks, using the UNSW-NB15 benchmark dataset as training data. The experimental results show that the proposed framework 111 14 achieves a maximum accuracy of 97.91% by outperforming the state-of-the-art methods and is 112 able to overcome the issues regarding the data noise and heterogeneity of IoT systems. The proposed research bears a significant generalizability score needed to enhance IoT security under the merged hyperparameter tuning approach and opens avenues for future work on model deep learning integration along with rigorous testing on all-inclusive datatypes.
Keywords: Cyber Security, Enhanced Bayesian Optimization, Hyperparameter Tuning, Internet of Things, Machine Learning.