Efficient Machine Learning for Prediction of Malicious URLs under Neutrosophic Uncertainty Framework

Abstract

With more than 5.44 billion users, the Internet is an essential component of everyday life, 
facilitating e-commerce, interaction, learning, and more. But with the proliferation of harmful 
Uniform Resource Locators (URLs), this widespread Internet access also raises questions about 
online security and privacy. Due to their significant advantages of lowering model variance, 
increasing predictive performance, raising prediction accuracy, and exhibiting strong 
generalization potential, traditional ensemble models have recently drawn interest. However, 
there is still work to be done on how to use it to combat rogue URLs. These URLs are dangerous 
to people and organizations because they frequently lurk behind static links in emails or web 
pages. Many malicious websites avoid detection despite blacklisting services because they are 
either newly created or not closely monitored. Hence, we use different machine learning (ML) 
models such as decision tree, AdaBoosting, Naïve Bayes, random forest, gradient boosting, and 
XGBoosting. Then these models are evaluated under the neutrosophic framework to deal with 
uncertainty. The WASPAS method is used to select the best ML model from different models. The 
results show that the random forest is the best ML model.

DOI: 10.5281/zenodo.15130967