A Novel Two-Step Bayesian Hyperparameter Optimization Strategy for DoS Attack Detection in IoT
Somnath Sinha, Shirasthi Choudhary, Aditi Paul · 2024
Variations of Hyperparameter in Machine Learning (ML) algorithm effectively strikes the model's performance in terms of accuracy, loss, F1 score and many others. In the current study a two-step hyperparameter optimization approach is represented to analyse selected ML models' performance in detecting specific Denial of Service attacks in IoT. These attacks are Synchronization Flooding Attack at Transport layer, DIS Flooding attack and Sinkhole attack at Network layer. The two-step approach is a combination of Manual Hyperparameter tuning followed by Bayesian Optimization technique. The first stage manually analyses the hyperparameters of ML algorithms by considering the nature of the attack datasets. This technique is quite rigorous as it demands thorough analysis of the dependencies of the nature of datasets with hyperparameter types. At the same time this process is time consuming. The output of the first stage is the ranges of independent hyperparameter values that give maximum accuracy (minimum error rate). In the next stage Bayesian Hyperparameter tuning is used to specifically derive the single set of all hyperparameters values that give optimized accuracy faster than the BO. The input to the second stage is the ranges of individual hyperparameters that gave maximum accuracy in the first stage. The efficiency of the approach is depicted by comparative analysis of training time between the proposed and existing BO. NetSim simulator is used for generating attack datasets and Python packages are used for executing the two-step approach.