Research Article
Performance Analysis of Boosting Techniques for Classificationand Detection of Malicious Websites
@INPROCEEDINGS{10.4108/eai.7-12-2021.2314506, author={Regis Anne W and CarolinJeeva S}, title={Performance Analysis of Boosting Techniques for Classificationand Detection of Malicious Websites }, proceedings={Proceedings of the First International Conference on Combinatorial and Optimization, ICCAP 2021, December 7-8 2021, Chennai, India}, publisher={EAI}, proceedings_a={ICCAP}, year={2021}, month={12}, keywords={malicious benign machine learning boosting cyber security lgbm xgboost and gradient boosting accuracy precision recall and support}, doi={10.4108/eai.7-12-2021.2314506} }- Regis Anne W
CarolinJeeva S
Year: 2021
Performance Analysis of Boosting Techniques for Classificationand Detection of Malicious Websites
ICCAP
EAI
DOI: 10.4108/eai.7-12-2021.2314506
Abstract
Phishing is a method of social engineering technique to deceive web users to capture sensitive information like user name and password in websites without the knowledge of the end user. The end user provides information about their personal and financial thinking it’s the authenticated service provider. URL meaning the "Uniform Resource Locator" that identifies an address to a file in the server. The URLs can be categorized as benign or malicious. Malicious URLs are created for the purpose of attacking to create loss and poses great threat to the victims. Machine Learning approaches offer a wide range of algorithms to detect malicious websites. It considers the URL as a set of features of Lexical, Host based and Content features to train a model to classify it as malicious or benign. Boosting is a collection of algorithms that combine the weaklearning classifiers to build strong Classifiers. In this paper boosting algorithms are exploited to the study of URL detection as malicious or benign. Boosting algorithms such as LGBM, XGBoost and Gradient Boosting are used for predicting phishing URL is presented. Feature selection to identify the important features is performed. The selected features are then classified by Random Forest Classifier to give an accuracy of 99%.
