Detecting Phishing Attacks Using Feature Importance-Based Machine Learning Approach

Phishing is a fraudulent technique that involves creating a malicious website or link to trick people into revealing sensitive information such as passwords, financial details, or personal information. It is a common form of cybercrime that can lead to identity theft and other harmful consequences. To address this serious threat, researchers have explored various approaches to detect it, including heuristic, list-based, and machine learning approaches. In our study, we developed an automated phishing detection approach that selects important features through feature selection based on the feature importance method. Once the important features were identified, we built models using Random Forest, KNN, and Logistic Regression algorithms. For our study, we utilized the ISCXURL-2016 dataset and employed a feature selection process to identify and select only significant features. By doing so, our approach achieves a remarkable accuracy of 98.85% with the Random Forest algorithm. Notably, the scalability of our method is enhanced because our method automatically extracts the important features from the data and eliminates the need for a human to select the features using domain knowledge and experience. As a result, our approach's scalability makes it a feasible solution for detecting phishing attacks in real-world scenarios.