Application of ensemble Machine Learning models for phishing detection on web networks
Navyah Puri, Pranay Saggar, Amandeep Kaur, Puneet Garg
Abstract
Phishing is a technique of fraud and identity stealing that includes convincing Website visitors to provide confidential info and details such as their user id, secret key, payment info, and so on. It is one of the real safety concerns that the online revolution is worried about, and it may cost businesses and users money. The use of SHAP values to better comprehend the model employed in phishing URL detection is the research’s highlight. This research examines multiple machine learning models for detecting phishing by examining various aspects of the website’s URL. The dataset that was used to train the model is open source, consisting of datasets from Alexa, UCI, Phishtank, and Kaggle. There are 11,055 rows and 32 columns in the data set. The data were normalized using the SMOTE analysis technique, which resulted in a larger data set. This data was then fed into a variety of classification and ensemble models (K-means, Random forest, decision tree, CatBoost classifier, LightGBM classifier, AdaBoost, and voting classifier). The Accuracy and F1 values of the models were compared. The model’s accuracy was tested before and after using smote. After putting all of the strategies to the test, we observed that CatBoost Classifier produced the best results for accuracy and F1 value. To conclude, SHAP values are a crucial part in model interpretation and are utilized to identify important features in the model and how they impact the output of the model. This model can be used by authorities and companies to stop phishing attacks and identify suspicious sites before someone is harmed by them.