This repository contains the data processing, model training, evaluation, and export pipelines for Bazarya, a lightweight market price forecasting and alert system designed for local markets in Ethiopia.
It serves as the ML backend for the Bazarya Flask app, which provides a web interface for price forecasts, user submissions, and alerts.
To forecast daily prices of agricultural and food products using time series models (e.g., ARIMA, LSTM, ). This repo ensures:
- Clean and structured market price data
- Automated preprocessing and feature engineering
- Reproducible training and evaluation of models
- Export of serialized models (e.g.,
.pkl,.h5) for use in production
- ✅ Support for multiple markets per product
- ✅ Support for categorical encodings
- 🔄 Multivariate models (e.g., inflation, weather)
- 🔄 Integration with real-time web scrapers
- 🔄 HuggingFace/DVC pipeline versioning
First Run the notebooks and export all the 240 models and save them in folder named "models" to use them in the Bazarya Flask app repo.
Trained models from this repo are loaded in the Bazarya Web App for live price prediction.
Ensure the model file paths in the Flask repo point to exported files from this repo.
Feel free to:
- Add new model types (e.g., Prophet, XGBoost)
- Improve preprocessing or scaling techniques
- Report issues or raise pull requests
Questions or feedback? email: Keneansufa@gmail.com