An Ensemble Method for Phishing Websites Detection Based on XGBoost
Jiaqi Gu, Hui Xu
Abstract
N owadays, the internet is spreading widely around the world. Beside the benefits the internet brings to its users, there are many potential harms such as phishing scams. These scams act like normal websites, stealing confidential information from users. To protect the privacy of the internet users from being imperiled by scams, it is necessary to find a way to detect those phishing websites. When considering detection, Machine Learning methods that have the best possible performances are promising techniques to resolve the problem since they can make reliable predictions or classifications for unseen scenarios upon given data. In this paper, I present a ensemble model to detect phishing websites using URL features. I used the dataset named “Phishing website Detector - phishing website dataset” from Kaggle. Next, several models using all types of mainstream ensemble algorithms (such as stacking, boosting, and bagging) were built. Finally, different methods measuring the performance of models were used. The reason why only ensemble models are selected as the methods to be implemented is mainly due to their overall strong performance. As for the results, the proposed XGBoost model combining Random Forest and K-Nearest Neighbors outperforms all other models (Random Forest, AdaBoost, Parameter-tuned XGBoost, Stacking, and Voting), achieving an accuracy of 99.74 % on training data and 96.44% on testing data. The confusion matrix generated according to the performance of the XGBoost model shows that it correctly predicts 1523 positive terms out of 1594 in total and 1995 negative terms out of 2054 in total. This resulting method can effectively detect phishing web sites. Furthermore, if integrated into application or web extension, it can protect internet users' privacy from being imperiled when they accidentally step into phishing web sites traps.