Understanding Phishing and Challenges in Detection Phishing involves creating replica websites to steal sensitive information from users. These fake sites mimic legitimate ones, making it difficult for individuals to distinguish between real and fraudulent platforms. The challenge lies in identifying phishing websites based on features rather than training systems with millions of website names, as this approach is impractical.
Proposed Solution Using Gradient Boosting Classifier The project introduces a machine learning model using gradient boosting classifier to detect phishing websites by analyzing specific features like URL length, IP address presence, SSL state, etc., instead of URLs themselves. Achieving 98.9% training accuracy and 97.4% test accuracy demonstrates its effectiveness compared to previous models that used algorithms like logistic regression or Naive Bayes but had lower performance metrics.
Implementation Process and Performance Analysis Users upload datasets containing feature records into the system which then predicts whether a site is safe or unsafe through an interface without relying on databases for dynamic updates. Real-time testing confirms accurate predictions against both legitimate (e.g., Amazon) and known phishing sites (e.g., fake SBI). Static charts derived from dataset classifications showcase consistent results across various scenarios with approximately 60% legitimate versus 40% phishing data distribution.