AI Phishing Detector

An AI-powered phishing detection system built using ensemble machine learning models. This project analyzes website metadata and predicts whether a URL is legitimate or a phishing attempt, with over 95% accuracy.

---

Overview

This project demonstrates the use of supervised machine learning to detect phishing websites using a dataset sourced from Kaggle. It uses two base models — Logistic Regression and Random Forest — combined into an ensemble classifier to increase accuracy and reliability.

---

Features

• Detects phishing vs. legitimate websites
• Ensemble model: Logistic Regression + Random Forest
• Pink-themed confusion matrix for visual analysis
• Scalable, modular codebase with clean function separation
• Accuracy score, classification report, and visual output

---

Technologies Used

• Python
• Pandas – data manipulation
• Scikit-learn – machine learning algorithms
• Matplotlib – data visualization
• VotingClassifier – for ensemble learning

---

Skills Demonstrated

• Building ensemble ML models from scratch
• Feature scaling and preprocessing techniques
• Model evaluation using confusion matrices and classification reports
• Structuring Python projects into clean, reusable functions
• Using .gitignore to manage sensitive/local files
• Version control with Git + GitHub

---

How It Works

1. Loads and preprocesses phishing dataset
2. Splits into training and testing sets
3. Scales feature data using StandardScaler
4. Trains an ensemble model (soft voting classifier)
5. Evaluates model with accuracy score, confusion matrix, and report
6. Displays results via terminal + pink-themed visual

Dataset used: Phishing Website Detector (Kaggle)

Note: To run this project, download phishing.csv from the link above and place it in the same directory as the Python script.

---

Results

• Final Accuracy: ~95.7%
• Classification Report: Includes precision, recall, F1-score
• Visual Analysis: Confusion matrix with styled visualization

---

Results Visualization

Confusion matrix for the model (pink-themed):

---

Future Improvements

• Deploy the model via Flask or Streamlit for web input
• Accept live URLs and perform real-time checks
• Experiment with deep learning or NLP for phishing email content
• Integrate threat intelligence feeds for smarter detection