dharshiyan/Multiclass-Spam-and-Phishing-Message-Detection-

InboxSentinel

Multiclass Spam & Phishing Message Detection using Natural Language Processing and Machine Learning

An end-to-end Machine Learning solution that classifies SMS and text messages into Ham, Spam, and Phishing using NLP techniques and Logistic Regression.

---

📌 Project Overview

This project implements a multiclass text classification system to classify messages into:

• Ham (Normal messages)
• Spam (Unwanted/Promotional messages)
• Phishing (Fraudulent/Scam messages)

The system uses Natural Language Processing (NLP) techniques and Machine Learning to preprocess text, extract features using TF-IDF, and classify messages using a Logistic Regression model. A Streamlit web application is also developed for real-time predictions.

---

✨ Project Highlights

• End-to-End NLP Pipeline
• Multiclass Classification
• 515,000+ Message Dataset
• TF-IDF Feature Engineering
• Logistic Regression Classifier
• 95.8% Classification Accuracy
• Streamlit Web Application
• Kaggle Dataset & Model Published

---

🎯 Problem Statement

Spam and phishing messages pose significant risks in communication systems. This project aims to build a robust machine learning model that can automatically detect and classify such messages to improve security and filtering systems.

---

🧠 Solution Approach

The project follows a complete NLP pipeline:

• Data collection from multiple sources
• Text preprocessing (cleaning, stopword removal)
• Feature extraction using TF-IDF
• Model training using Logistic Regression
• Model evaluation using classification metrics
• Model deployment using Streamlit

---

⚙️ Machine Learning Pipeline

Raw Data → Preprocessing → TF-IDF → Model Training → Evaluation → Model Saving → Deployment

---

🚀 Key Features

✔ Text Cleaning

✔ Tokenization

✔ Stopword Removal

✔ TF-IDF Vectorization

✔ Multiclass Classification

✔ Real-time Prediction

✔ Model Serialization

✔ Interactive Streamlit Interface ---

📊 Model Performance

• Accuracy: ~95.8%

| Class | Precision | Recall | F1-Score | | -------- | --------- | ------ | -------- | | Ham | 0.99 | 0.98 | 0.98 | | Spam | 0.97 | 0.96 | 0.97 | | Phishing | 0.87 | 0.90 | 0.88 |

The model achieves strong performance across all classes with effective phishing detection.

---

🔗 Project Resources (Kaggle Integration)

• 📊 Dataset:

https://kaggle.com/My_Dataset-Link

• 🤖 Trained Model:

https://kaggle.com/My_Model-Link

• 📒 Training Notebook:

https://kaggle.com/My-Notebook-Link

These resources ensure full reproducibility of the project.

---

📊 Dataset

• Dataset Size: 515,000+ labelled messages
• Classes: 3
• Ham
• Spam
• Phishing

---

🛠 Technologies Used

Programming Language

• Python

Machine Learning

• Scikit-learn
• Logistic Regression

Natural Language Processing

• NLTK
• TF-IDF Vectorizer

Deployment

• Streamlit

Utilities

• Pandas
• NumPy
• Pickle

---

📁 Project Structure

spam-phishing-detection/
│
├── app.py
├── train.py
├── requirements.txt
├── README.md
│
├── models/
│   ├── model.pkl
│   └── vectorizer.pkl
│
├── data/
│   └── dataset_sample.csv
│
├── assets/
│   └── confusion_matrix.png
│
└── .gitignore

---

🚀 How to Run the Project

1️⃣ Clone the Repository

git clone https://github.com/dharshiyan/spam-phishing-detection.git
cd spam-phishing-detection

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Run the Application

streamlit run app.py

---

💡 Example Predictions

| Message | Prediction | | ----------------------------------------------- | ---------- | | "Congratulations! You won a free iPhone" | Spam | | "Let's meet tomorrow at 5 PM" | Ham | | "Verify your bank account password immediately" | Phishing |

---

📈 Results

The final Logistic Regression model achieved excellent performance on the test dataset with an overall accuracy of 95.8%, demonstrating strong generalization capability across all three message categories.

The model successfully distinguishes phishing attempts from legitimate messages while maintaining high precision and recall. ---

📈 Future Improvements

• Implement deep learning models (LSTM / BERT)
• Add URL-based phishing detection features
• Improve dataset balance
• Deploy API using Flask or FastAPI
• Dockerize the application

---

👨‍💻 Author

Developed by

DHARSHIYAN

Artificial Intelligence & Data Science Undergraduate

Machine Learning | Deep Learning | Computer Vision | NLP

---

📜 License

This project is licensed under the MIT License.