Abstract:

Abstract In recent years, the rapid growth of cyberspace has led to an increase in challenges related to information security. One of the most dangerous cyberattacks is website phishing, which is complex in nature and difficult to detect in real-time. Cloud Machine Learning has emerged as an effective approach for detecting website phishing by leveraging Cloud Computing Services to obtain accurate results quickly. Therefore, this study presents a Cloud Machine Learning method for evaluating and assessing the time required to detect website phishing using three SageMaker built-in algorithms: Extreme Gradient Boosting, Linear Learner, and k-Nearest Neighbor. Amazon Web Services is utilized for storage, training, evaluation, and online deployment over a large dataset of 11,430 samples and 89 features. The results indicate that Extreme Gradient Boosting outperformed the other two algorithms with an accuracy of 96.4% and an online prediction time of 0.0005 minutes, followed by Linear Learner with an accuracy of 94.4% and a prediction time of 0.0006 minutes. While k-Nearest Neighbor obtained the lowest accuracy score of 83.7% and the longest prediction time of 0.0008 minutes.

Keywords: