Scam Detection Using NLP & Machine Learning

Overview

This project is a Scam Detection System designed to identify fraudulent messages across text-based communication platforms like (e.g., WhatsApp, Instagram, Telegram). It uses a hybrid approach combining Natural Language Processing (NLP) techniques (TF-IDF, N-gram analysis) and Machine Learning (Random Forest) to detect scams with high accuracy and minimal false positives.

Features

• Hybrid Detection: Combines pattern-based (N-gram) and ML-based (TF-IDF + Random Forest) analysis.
• Multi-Platform: Detects scams from various messaging platforms.
• Scam Type Classification: Identifies the type of scam (phishing, impersonation, financial, urgent, etc.).
• Continuous Learning: Learns from user feedback to adapt to new scam tactics.
• Educational Component: Provides information to help users recognize scams.

Project Structure

• run.py — Main entry point; starts the Flask web server and runs evaluation.
• scam_detector.py — Core logic for text processing, feature extraction, and ML models.
• routes.py — Flask routes for web/API interaction.
• index.html — Web interface for user interaction.
• __init__.py — Flask app initialization.
• Financial.html — Educational page for financial scams.
• Phishing.html — Educational page for phishing scams.
• Urgency.html — Educational page for urgent scams.
• Impersonation.html — Educational page for impersonation scams.

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/Scam-Detection-Using-NLP-Machine-Learning-Text-Based.git
   cd Scam-Detection-Using-NLP-Machine-Learning-Text-Based

2. Create and activate a virtual environment (recommended):

   python3 -m venv venv
   source venv/bin/activate

3. Install dependencies:

   pip install flask scikit-learn pandas numpy nltk

4. Download NLTK resources (automatically handled on first run):

   • The system will download required NLTK resources (punkt, stopwords, wordnet, averaged_perceptron_tagger) if not already present.

Usage

1. Run the application:

   python run.py

2. Access the web interface:

   • Open your browser and go to: http://127.0.0.1:5000

3. How it works:

   • Paste or type a message into the web interface.
   • The system will analyze the message and display the risk level, scam type, and important features.
   • You can provide feedback to help the system learn and improve.

Evaluation

• On startup, the system performs a comprehensive evaluation (cross-validation and holdout test) and prints results in the terminal.
• Metrics include accuracy, precision, recall, F1 score, and a confusion matrix.

Notes

• This is a Web Application (Not directly implemented into WhatsApp or Instagram.
• The system currently supports only English-language messages.
• Training data is stored in data/training_data.csv and grows as users provide feedback.
• In the event of dataset corruption, the system incorporates a backup mechanism that restores the dataset using the flagged_csv file. (This feature was implemented to address issues encountered during the production phase of the project)

License

MIT License