Machine Learning for Real-Time SMS Spam Detection in Cellular Networks

Overview

This project evaluates the effectiveness of multiple machine learning models for real-time spam detection in cellular networks. Using a comprehensive SMS dataset, we train, evaluate, and simulate four different ML models in a realistic cellular network environment to determine the most effective approach for real-time spam filtering. View the academic report here.

Key Features

• 4 Machine Learning Models: Logistic Regression, Naive Bayes, Random Forest, and Support Vector Machine
• Real-time Simulation: Simulated cellular network environment with baseband units and radio units
• Comprehensive Evaluation: Performance metrics including accuracy, precision, recall, and F1-score
• Automated Spam Detection: Real-time alerting system for high spam volume detection
• Dataset Generation: Custom spam dataset generator for testing on unseen data

Research Focus

The project addresses the critical need for effective spam detection in cellular networks by:

• Comparing multiple ML algorithms in a realistic network simulation
• Evaluating real-time performance under cellular network constraints
• Analyzing model effectiveness for different spam patterns and volumes
• Providing insights into the most suitable algorithms for mobile network deployment

Model Performance Summary

Model	Training Accuracy	Simulation Accuracy	Simulation Time	Best Use Case
Logistic Regression	99%	88%	8m 47s	High precision spam detection
Naive Bayes	89%	81%	9m 39s	Fast processing, good recall
Random Forest	89%	85%	20m 55s	Balanced performance
Support Vector Machine	89%	82%	25m 53s	Complex pattern recognition

Project Structure

machine-learning-for-spam-sms/
├── 📁 models/                          # ML Model Development
│   ├── models.ipynb                    # Model training & evaluation
│   ├── spam_data.csv                   # Training dataset
│   └── *.pkl                          # Trained model files
├── 📁 simulation/                      # Network Simulation
│   ├── simulation.ipynb                # Main simulation notebook
│   ├── 📁 data/                       # Generated test datasets
│   ├── 📁 logs/                       # Simulation logs by model
│   └── 📁 results/                    # Performance results & figures
├── 📁 spam-generator/                  # Dataset Generation
│   ├── generator.py                    # Spam dataset generator
│   └── conversations.py               # Conversation templates
├── 📁 markdown/                        # Documentation
│   ├── models.md                       # Model implementation details
│   ├── simulation.md                   # Simulation methodology
│   └── install_instructions.md         # Setup instructions
└── requirements.txt                    # Python dependencies

Key Components

• Machine Learning Models: Four different algorithms trained on SMS spam data
• Cellular Network Simulation: Realistic network topology with baseband and radio units
• Real-time Processing: Stream processing of SMS messages with spam detection
• Performance Monitoring: Comprehensive logging and alerting system
• Dataset Generation: Custom spam generator for testing model robustness

Quick Start

Prerequisites

• Python 3.8 or higher
• Virtual environment (recommended)

Installation

1. Clone the repository

   git clone https://github.com/breezy-codes/machine-learning-for-spam-sms.git
   cd machine-learning-for-spam-sms

2. Set up virtual environment

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .\.venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

For detailed setup instructions, see: Setting Up a Python Virtual Environment

🤖 Running the Machine Learning Models

Train and evaluate all four ML models using the comprehensive Jupyter notebook:

jupyter notebook models/models.ipynb

What the Models Do

• Data Preprocessing: Text cleaning, tokenization, and vectorization
• Model Training: Hyperparameter tuning with cross-validation
• Performance Evaluation: Accuracy, precision, recall, F1-score metrics
• Model Persistence: Saves trained models as .pkl files

Detailed guide: Model Implementation Notes

Running the Cellular Network Simulation

Experience real-time spam detection in a simulated cellular environment:

jupyter notebook simulation/simulation.ipynb

Simulation Features

• Network Topology: Multiple baseband units with radio units
• Real-time Processing: Stream-based message processing
• Spam Detection: Live classification with alerting system
• Performance Analytics: Comprehensive logging and metrics collection
• Load Testing: Handles high-volume message streams

Detailed guide: Simulation Methodology

Results & Analysis

Model Performance Comparison

The simulation reveals interesting trade-offs between different algorithms:

• Logistic Regression: Highest precision (99% spam detection) but lower recall
• Random Forest: Best balanced performance with 85% accuracy
• Naive Bayes: Fastest processing with good spam recall (90%)
• SVM: Robust to outliers but computationally intensive

Real-time Performance Insights

• Processing Speed: Naive Bayes processes messages fastest
• Memory Usage: Logistic Regression has smallest memory footprint
• Accuracy vs Speed: Random Forest offers best accuracy/speed balance
• Alert Response: All models successfully trigger spam volume alerts

🛠️ Technical Architecture

Machine Learning Pipeline

1. Data Preprocessing: Text normalization, stop word removal, stemming
2. Feature Extraction: TF-IDF vectorization with n-grams
3. Model Training: Cross-validation with hyperparameter optimization
4. Evaluation: Multi-metric assessment on held-out test data

Cellular Network Simulation

┌─────────────────┐     ┌──────────────────┐     ┌─────────────────┐
│   Radio Unit    │───▶│  Baseband Unit   │───▶│  Core Network   │
│  (Message RX)   │     │  (ML Processing) │     │ (Spam Alerts)   │
└─────────────────┘     └──────────────────┘     └─────────────────┘

• Radio Units: Simulate message reception from mobile devices
• Baseband Units: Apply ML models for real-time spam classification
• Core Network: Aggregate results and trigger spam volume alerts
• Logging System: Captures all decisions and performance metrics

Customization & Extension

Adding New Models

1. Train your model in models/models.ipynb
2. Save as .pkl file in the models/ directory
3. Add simulation code in simulation/simulation.ipynb
4. Update logging and results directories

Modifying Network Topology

• Adjust baseband unit count in simulation parameters
• Configure radio unit connections per baseband
• Customize message processing rates and volumes

Custom Dataset Generation

Use the spam generator to create targeted test scenarios:

from spam_generator.generator import generate_spam_dataset
dataset = generate_spam_dataset(volume=1000, spam_ratio=0.3)

Dependencies

Key libraries used in this project:

• scikit-learn: Machine learning algorithms and evaluation
• pandas: Data manipulation and analysis
• numpy: Numerical computing
• matplotlib/seaborn: Data visualization
• nltk: Natural language processing
• simpy: Discrete event simulation
• jupyter: Interactive development environment

Contributing

Contributions are welcome! Areas for improvement:

• Additional ML algorithms (Deep Learning, XGBoost)
• Enhanced network simulation (5G features, edge computing)
• Real-world dataset integration
• Performance optimization
• Mobile deployment strategies

License

This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.

References

• SMS Spam Collection Dataset
• Cellular Network Architecture Standards
• Machine Learning for Telecommunications
• Real-time Stream Processing Techniques

---

Built with ❤️ for telecommunications and machine learning research