Swahili Spam Detection - Training Repository

The training component of the Swahili Spam Detection system, containing datasets, training notebooks, and model evaluation infrastructure. This repository handles the machine learning aspects of the main SSD project.

Project Structure

└── ssd-training/
    └── dataset/             # Training datasets
        └── combined_set.csv
        └── combined_set.xlsx
    └── Model_After_Training/
        └── model_after_tranining_28_jan_2025/
            └── swahiliSpamDetectionModel.pkl
        └── swahiliSpamDetectionModel.pkl
    └── stopwords/           # Swahili stopwords
        └── Common Swahili Stop-words.csv
    └── spamDetectionRef.ipynb   # Training notebook
    └── requirements.txt         # Dependencies

Model Performance

Performance Metrics

Model	Technique	Accuracy	Precision	Recall	F1-score	AUC-ROC
Logistic Regression	Count Vectorization	0.9924	0.9925	0.9924	0.9923	0.9996
Naive Bayes	Count Vectorization	0.9904	0.9905	0.9904	0.9905	0.9988
SVM	Count Vectorization	0.9933	0.9934	0.9933	0.9933	0.9995
Random Forest	Count Vectorization	0.9933	0.9933	0.9933	0.9933	0.9993
Logistic Regression	TF-IDF	0.9838	0.9842	0.9838	0.9837	0.9995
Naive Bayes	TF-IDF	0.9952	0.9952	0.9952	0.9952	0.9985
SVM	TF-IDF	0.9914	0.9915	0.9914	0.9914	0.9999
Random Forest	TF-IDF	0.9924	0.9925	0.9924	0.9923	0.9998

Getting Started

Prerequisites

• Python 3.8+
• pip package manager

Installation

1. Clone the repository

git clone https://github.com/patrick-paul/ssd-training.git
cd ssd-training

2. Create virtual environment

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

3. Install dependencies

pip install -r requirements.txt

Dataset Information

• Contains over 6,000 labeled Swahili messages
• Binary classification (spam/ham)
• Available in CSV and Excel formats
• Includes regional language variations

Model Details

Best Performing Model

Random Forest with Count Vectorization:

• Accuracy: 0.9933
• Precision: 0.9933
• Recall: 0.9933
• F1-score: 0.9933
• AUC-ROC: 0.9993

Key Features

• Robust against overfitting
• Consistent performance across metrics
• Excellent handling of text features
• Strong performance with Count Vectorization

Dependencies

pandas
matplotlib
numpy
seaborn
nltk
scikit-learn
joblib

Usage

1. Open the Jupyter notebook:

jupyter notebook spamDetectionRef.ipynb

2. Follow the notebook sections:
   • Data preprocessing
   • Feature extraction
   • Model training
   • Performance evaluation
   • Model export

Contributing

1. Fork the repository
2. Create your feature branch
3. Follow PEP-8 standards
4. Push your changes
5. Submit a pull request

Future Improvements

• Expand dataset with more regional dialects
• Implement deep learning models
• Add automated model retraining pipeline
• Create comprehensive model testing suite
• Add model versioning system

License

MIT License - See LICENSE.md for details

Contact

• Developer: [email protected]
• GitHub: @patrick-paul
• Project Link: https://github.com/patrick-paul/ssd-training

Related Projects

• Main SSD Application - The web application using these trained models