Detection of adversarial phishing attack using machine learning techniques

The frequency of cyberattacks, particularly phishing attacks, is increasing exponentially every day. Many users fall victim to clicking on malicious URLs, leading to the exploitation of their information. The traditional methodologies employed to prevent such attacks are failing to detect adversarial malicious URLs. Common techniques, such as comparing URLs with a bag of words and linguistic features, are inadequate in preventing adversarial attacks. Our model extracts web-scraped features, utilizing 89 features for accurate phishing URL classification. Based on feature selection algorithms like forward selection and lasso regularization, we have chosen the best 27 features. We employ ensemble learning algorithm such as Random Forest, AdaBoost, GradientBoost, and XGBoost algorithm to assess its effectiveness, utilizing various evaluation metrics. Additionally, we have developed a webpage for users to enter a URL and determine whether it is legitimate or not. This phishing detection is done based on the model pickle file generated during model building.