Comparison of Ensemble Models for the classification of Malicious URLs
Ngaira Mandela, Amir Shaker, Felix Etyang
Abstract
Abstract: The increasing number of malicious URLs on the internet poses a significant threat to online security and privacy of individuals and organizations. Machine learning algorithms have been proposed as a solution to this problem, but the high volume and diversity of URLs and the constant evolution of malicious URLs pose significant challenges for researchers in this field. Ensemble models, which combine multiple models to improve the overall performance, have shown great promise in addressing these challenges. In this paper, we compare four popular ensemble algorithms for the task of classifying URLs: Random Forest, Bagging, XGBoost and AdaBoost. We use a dataset of labeled URLs and pre-process them. The performance of the models is evaluated using accuracy and run time as the evaluation metric. Our results show that all the four algorithms are able to achieve good performance with accuracy scores above 95%. The Random Forest algorithm had the highest accuracy, followed by XGBoos, Bagging and then AdaBoost. XGBoost was the fasted algorithm with runtime of 1 minute. Our study provides insight into the relative strengths of these ensemble algorithms on the task of URL classification and highlights the importance of selecting an appropriate model depending on the specific characteristics of the data and the requirements of the application.