Detecting Phising Website using Machine Learning Methods

Modern phishing is a sophisticated cyberattack. It steals login details and payment information by posing as a trusted entity. With our increasing dependence on digital platforms, the frequency and sophistication of phishing attacks have escalated. Many studies have examined phishing detection. Most prioritize metrics like accuracy, precision, and recall. However, a significant research gap exists as previous studies have primarily focused on accuracy and evaluation metrics without incorporating validation processes to verify whether the developed models are adequately robust and reliable. Harnessing the power of machine learning, we can classify websites into authentic or fraudulent categories, proposing a robust defense against these malicious schemes. The purpose of this study is to present the novelty of a concise summary of techniques for detecting phishing and to create a framework that can be used to detect phishing. The dataset comprises 662,590 entries and 9 features. This study implements three supervised learning models: Decision Tree, K-Nearest Neighbors (KNN), and XGBoost algorithms. These algorithms were chosen for their dataset knowledge and applicability. According to experiments, the Decision Tree model has the lowest accuracy at 88.66% and the KNN model the highest at 88.94%. The XGBoost model records an accuracy of 90.24%. XGBoost often achieves high accuracy due to its gradient boosting framework, which combines multiple decision trees to minimize errors. Its regularization techniques also help prevent overfitting, leading to robust performance on unseen data.