Multiple-Multivariate Time Series Forecasting
End-to-End Python based ML project focusing on forecasting multiple multivariate time series with production grade deployment techniques.
Overview
This is a Python, ML-based API with an impressive Normalised Mean Squared Error of 0.039 and Normalised Mean Absolute Error of 0.10 for forecasts of this multiple, multivariate timeseries data on a 15-day window. This POC aims to empowers retailers with data-driven sales forecasting to optimize inventory management and improve profitability. This comprehensive solution predicts short-term sales (up to 30 days) with outstanding accuracy and performance metrics. By integrating seamlessly with existing workflows, it enables retailers to:
- Make informed inventory decisions.
- Reduce stockouts and overstocking.
- Improve operational efficiency.
- Boost customer satisfaction.
Tech Stack
- Darts : For efficient time series operations and forecasting.
- MongoDB : For storage and retrieval of data.
- LightGBM : To accurately predict covariate and target features.
- Scikit-learn : For creating data pipelines.
- DVC, Git, and Github : For seamless data and code versioning.
- Evidently AI: To check for data drift/target drift.
- FastAPI : For building a user-friendly API for model accessibility.
- Docker and Dockerhub : For secure and streamlined deployment to the Render Cloud Platform.
- Github Actions : For automating the CI/CD pipeline.
- Render Cloud : A PaaS for deployment of web apps, api’s, etc.