Lightweight Hyperparameter Optimization Model for Enhancing Phishing Detection in IoT

This study presents an enhanced machine learning approach that emphasizes the optimization of hyperparameters to improve phishing detection, particularly in resource-constrained environments like Internet of Things (IoT) devices. Phishing is considered one of the dangerous forms of cyberattacks where attackers can reveal sensitive information about user's identity, password, privacy and even properties. Machine learning techniques and tools are playing important role in detecting phishing and have shown to be effective and advantageous methods for detection and classification, especially for the unified resource locator (URL). The proposed model presupposes a systematic approach for feature selection as well as finding the optimized hyperparameter values for the sake of increasing the detection quality while maintaining low computational complexity of the process. This study examines how feature set selection from a training dataset and how hyperparameters tuning can significantly improves the performance of phishing attack classification in IoT devices. Logistic regression, random forest, gradient boosting, support vector machine, and k-nearest neighbors are used in this study. According to the experimental, we found the best hyperparameter values for each classifier and comparative results of the implemented classification algorithms showed that support vector machine achieved the best performance with an accuracy of 96.2%.