Efficient And Accurate Technique For Improving ML Classifier Performance Using Feature Selection In Phishing Website Prediction

Abstract

Due to cyberattacks and various strategies, phishing websites are a problem on the Internet. One of these cyberattacks is phishing, in which the attacker pretends to be a trusted party to get sensitive and confidential information. Blacklisting, heuristic search, and visual similarity are just a few of the anti-phishing strategies that have been used to identify fraudulent activity. Machine learning (ML) techniques appear to be a beacon in the gloom of phishing websites, in contrast to these traditional methods, which take a long time to detect and have a high false rate. By introducing a novel features selection method in this article, it is possible to extract highly correlated features from datasets, thereby increasing the accuracy of classifiers over all features. Eight classifiers—Support vector machine (SVM kernel linear and rbf), Logistic regression (LR), Random forest (RF), Adaboost, Decision tree (DT), K-nearest neighbor (k-NN), and Gradient boosting (GBC)—as well as six feature selection techniques (Pearson, Chi-2, RFE, Logistics, Random Forest, and LightGBM) are used on phishing dataset with all features and feature selection methods. Comparing the results, it came to the conclusion that the random forest classifier and feature selection using the Chi-2 method have the potential to improve the model's accuracy. The accuracy of the proposed model reached as high as 96.99%.