Phishing continues to be a prevalent threat to the integrity of online banking platforms, exploiting user trust through deceptive web interfaces and fraudulent URLs. These attacks compromise sensitive information such as login credentials and financial data. In response, this study was initiated to develop an enhanced security model that not only detects phishing attempts but also prevents unauthorized access using cryptographic authentication. This paper aims to secure online banking platforms using a dual-layered approach combining machine learning with Visual Cryptography. To achieve this, a hybrid phishing detection and prevention system was designed and successfully implemented. The system integrates two core modules: an intelligent phishing detection engine and a secure authentication mechanism. The phishing detection engine combines K-Nearest Neighbors (KNN) for analyzing URL-based features with a Convolutional Neural Network (CNN) for image-based classification of websites. For authentication, the system generates two Visual Cryptographic (VC) shares per user during registration. One share is emailed to the user, while the other is stored securely on the server, enabling share recombination at login to verify identity. The solution was integrated with WordPress via REST API endpoints and tested extensively using both browser-based interactions and Postman. The system achieved 94% accuracy with the KNN model and 84% with the CNN model. However, our dual-model approach improves robustness and reduces reliance on one detection path. The average response time for model predictions was approximately 0.136 seconds on Render-hosted API, demonstrating reasonable computational efficiency for real-time use.
Keywords: Online Banking, Phishing, Security, k-nearest neighbor, visual cryptography
