Advancements in Phishing Website Detection: A Comprehensive Analysis of Machine Learning and Deep Learning Models

Today, machine learning (ML) and deep learning (DL) have become potent tools for detecting phishing websites. This paper has employed a bunch of ML algorithms such as Random Forest, Extra Tree, XGBoost, LGBM, and XGBoost-bagging, along with some proposed neural network architectures found on Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) for identifying website phishing. Implementing a comprehensive approach, we devised our model based on 56 proposed features extracted from each target website using Feature Importance. Then, we conducted five strategic steps to enhance the performance of ML and DL models. Through these steps, we improved DL model accuracy remarkably, notably elevating LSTM accuracy from the lowest accuracy of $96.18 \%$ to $98.78 \%$ with an AUC-ROC LSTM of about $\mathbf{9 9 . 8 5 \%}$.