🛡️ Phishing Website Prediction Model

This project uses Machine Learning and Natural Language Processing (NLP) techniques to detect phishing websites based on URL analysis. It aims to help users stay protected from cyber threats by enabling real-time and automated phishing detection.

---

📘 Abstract

This solution classifies website URLs as either legitimate ("good") or malicious ("bad"). By leveraging supervised learning models such as Logistic Regression and Multinomial Naive Bayes, we have built a pipeline that delivers high accuracy and can be deployed for real-time detection tasks.

---

🎯 Objective

To build an automated and scalable system that can classify phishing websites using:

• NLP techniques for URL feature extraction
• Supervised ML algorithms for classification
• Modular pipelines for deployment and interpretation

---

🧠 Motivation

Phishing attacks are increasingly sophisticated and often bypass traditional blacklist-based systems. Manual detection is slow and inaccurate, motivating the need for an intelligent, real-time solution.

---

❌ Existing Problems

• Dependence on outdated blacklists
• High false-positive rates
• Inability to detect new phishing patterns quickly

---

✅ Proposed Solution

• ML-based classification using Logistic Regression and Multinomial Naive Bayes
• Preprocessing steps: Tokenization, Stemming, Feature Extraction
• Visualizations: Word Clouds, Confusion Matrix, Performance Metrics
• Deployment-ready pipeline using pickle

---

🛠️ Technologies Used

Category	Tools & Libraries
Programming	Python
NLP	NLTK (tokenization, stemming)
ML	scikit-learn (vectorization, modeling)
Visualization	matplotlib, seaborn, WordCloud
Persistence	pickle
Platform	Jupyter Notebook / any Python IDE

---

🧱 System Architecture

1. Data Preprocessing

   • Clean and tokenize URLs
   • Apply stemming
   • Convert text to vectors (Bag-of-Words / TF-IDF)

2. Model Development

   • Train using Logistic Regression and Naive Bayes
   • Create a complete ML pipeline

3. Evaluation

   • Metrics: Accuracy, Precision, Recall, F1-Score
   • Visualizations: Word Clouds, Confusion Matrix

4. Deployment

   • Save model using pickle
   • Load and use for real-time predictions

---

📊 Results

Metric	Logistic Regression	Multinomial Naive Bayes
Accuracy	96.4%	95.8%
Precision, Recall, F1	High across all	High across all

• Word clouds reveal frequent tokens in good vs. bad URLs
• Clear visual output using matplotlib and seaborn

---

🔚 Conclusion

• This project presents a robust, automated phishing detection model
• Strong performance with basic ML models and URL-based features
• Demonstrates the power of NLP + ML in cyber threat detection

🚧 Limitations

• Heavily dependent on labeled training data
• May underperform on unseen or highly obfuscated URLs

🔮 Future Work

• Expand dataset diversity
• Try deep learning models (e.g., LSTMs, CNNs on text)
• Build a real-time API for production use

---

📁 Repository Structure

File	Description
presentation.pptx	PowerPoint summarizing the entire project
phishing_model.pkl	Saved pipeline model (if available)
notebook.ipynb	Jupyter Notebook with code and outputs
requirements.txt	Python dependencies

---

👥 Authors

• Sparsh Garg [RA2211003030192]
• Arpit Singhal [RA2211003030197]

Submitted to: Ms. Madhuri Sharma

---

📬 Contact

Feel free to connect or collaborate:

📧 Email: arpitsinghal00001@gmail.com
🔗 LinkedIn: Arpit Singhal

---

🛡️ Protecting users, one URL at a time.