kennethmugo/Swahili-SMS-Spam-Detection

# Swahili SMS Spam Detection ## Project Overview This project implements an end-to-end machine learning pipeline for detecting spam in Swahili text messages. It uses the LaBSE (Language-agnostic BERT Sentence Embeddings) model for text embedding and a binary classifier for spam detection. ### Key Features - Multilingual text processing using LaBSE - FastAPI-based REST API for real-time predictions - MLflow for experiment tracking and model versioning - DVC for data and model versioning - Docker support for containerized deployment - Modular architecture following clean code principles ## Getting Started ### Prerequisites - Python 3.10 - Conda (recommended for environment management) - Git LFS (for model files) ### Installation 1. Clone the repository: ```bash git clone github.com cd Swahili-SMS-Spam-Detection ``` 2. Create and activate conda environment: ```bash conda create -n swahili_spam python=3.10 conda activate swahili_spam ``` 3. Install dependencies: ```bash pip install -r requirements.txt ``` ### Running the Application 1. **Local Development**: ```bash uvicorn app:app --reload ``` 2. **Docker Deployment**: ```bash docker build -t swahili-spam-detector . docker run -p 8000:8000 swahili-spam-detector ``` 3. **View MLFlow Experiments**: ```bash mlflow server --host 127.0.0.1 --port 8080 ``` ## Technical Architecture ### Components 1. **Data Processing Pipeline** - Handles data ingestion and preprocessing - Generates embeddings using LaBSE model - Manages train-test splitting 2. **Model Pipeline** - Trains binary classifier on embeddings - Performs model evaluation - Logs metrics and artifacts to MLflow 3. **API Service** - FastAPI-based REST endpoint - Real-time prediction serving - Model loading and management ### Directory Structure ``` ├── artifacts/ # Generated artifacts (data, models) ├── config/ # Configuration files ├── model/ # Saved models ├── note …