cd ../portfolio
case study · machine learning · full-stack

Prophet Forecasting App

A full-stack web application for advanced time-series forecasting with Facebook Prophet — upload historical data, tune the model, and read the future through interactive charts.

Prophet Forecasting App — forecast chart illustration

01/ Overview

The Prophet Forecasting App makes advanced forecasting accessible to non-technical users. Upload an Excel file with historical data and generate accurate predictions with interactive visualizations and fully customizable model parameters — no notebooks, no code.

02/ Key features

  • Excel upload: .xls / .xlsx support with a drag-and-drop interface
  • Advanced forecasting: Facebook Prophet with multiple fallback methods
  • Interactive charts: Chart.js visualizations of history + predictions
  • Configurable parameters: seasonality, trend flexibility, holiday effects
  • Accuracy metrics: MAE, RMSE, and MAPE performance indicators
  • Holiday support: built-in holiday calendars for multiple countries
  • Responsive design: works seamlessly on desktop and mobile

03/ Technology stack

  • Frontend: React 18 + TypeScript, Chart.js, React Dropzone, mobile-first CSS
  • Backend: FastAPI, Facebook Prophet, Pandas & NumPy, layered fallback forecasting
  • Deployment: GitHub Pages (frontend), Render (API), GitHub Actions CI/CD

04/ Data contract

The app expects a simple, Prophet-native structure:

  • ds — date column in YYYY-MM-DD format
  • y — numeric values to forecast

05/ Use cases

  • Sales forecasting and revenue prediction
  • Demand planning and inventory management
  • Website traffic and user-growth analysis
  • Financial planning and budget forecasting
  • Research and academic time-series analysis

This project demonstrates full-stack engineering, applied data science, and modern deployment practice in one package — powerful machine learning behind an interface anyone can use.