Phishing Detection and Classification Using Machine Learning Algorithms

Project Synopsis & Overview

Phishing is a prevalent form of cybercrime in which attackers mimic legitimate websites to deceive users into revealing sensitive information such as passwords and financial data. Traditional detection methods—blacklist-based, heuristic-based, and content-based—struggle against zero-day attacks, suffer high false-positive rates, and require heavy manual rule maintenance. This project proposes an intelligent, data-driven phishing detection system that uses a Deep Neural Network (DNN) alongside a logistic regression baseline to classify websites as legitimate or phishing. The system is trained on a phishing dataset containing URL-based and content-based website features such as URL length, HTTPS presence, number of subdomains, domain age, and traffic. Preprocessing includes data cleaning, feature scaling with StandardScaler, and feature selection using SelectKBest with Mutual Information scoring to identify the ten most relevant predictors; class imbalance is addressed using SMOTE. The major modules are Data Collection, Preprocessing Layer, Model Training Layer, Prediction and Inference Layer, and Model Storage Layer, with the trained model saved in .h5/.pkl format for deployment. The workflow moves from raw feature ingestion through cleaning, scaling, feature selection, DNN training, and real-time inference. The expected output is a high-accuracy classifier suitable for integration into browser security extensions, network monitoring tools, and other real-time cybersecurity applications.

Complete Technical Specifications