Export Forecasting using ARIMA and LSTM

A full-stack Flask web application for export time series forecasting using ARIMA, LSTM, and a Hybrid ARIMA+LSTM model.
Users can register/login, upload monthly export CSV files, run forecasts, view interactive charts (Chart.js), and inspect forecast history.

Features

• User Authentication

  • Register/login/logout using bcrypt-hashed passwords
  • Server-side sessions (Flask-Session)

• Dataset Management

  • Upload CSV files (max 5MB) with a date column and one numeric series
  • Automatic date column detection and basic validation
  • Stores metadata: row count and date range

• Forecasting Models

  • ARIMA: classical time series forecasting
  • LSTM: recurrent neural network for sequence prediction
  • Hybrid: average of ARIMA and LSTM forecasts and metrics

• Metrics & Visualisation

  • MAE, RMSE, MAPE, and Accuracy
  • Main Chart.js line chart with historical vs forecast values
  • Model comparison view with RMSE bar chart

• Forecast History

  • Stores latest forecasts per user in SQLite
  • History table with model, dataset, horizon, and metrics
  • Reload forecast chart from stored JSON
  • Delete individual history records

Project Structure

See the export_forecasting/ structure in this repository (matches the specification: app.py, wsgi.py, database/, auth/, model/, templates/, static/, uploads/, saved_models/, etc.).

Local Setup

1. Clone and create virtual environment (recommended):

cd export_forecasting
python -m venv venv
venv\Scripts\activate  # on Windows

2. Install dependencies:

pip install -r requirements.txt

3. Run the app locally:

python wsgi.py

4. Open the app in your browser:

• Navigate to http://localhost:5000
• Register a new user, login, upload CSVs, and start forecasting.

Deploying on Render.com

1. Push to GitHub

   • Create a new GitHub repository and push this project.

2. Create a new Web Service

   • Go to Render and click New → Web Service.
   • Connect your GitHub repository.

3. Configure build and start commands

   • Build command:

     pip install -r requirements.txt

   • Start command:

     gunicorn wsgi:app

   (These are already set in render.yaml, which Render can auto-detect.)

4. Environment variables

   • Add SECRET_KEY (any strong random string) in the Render Environment tab, or let Render generate it based on render.yaml.

5. Deploy

   • Click Create Web Service. Render will install dependencies, build, and start the app.
   • Once live, open the public URL and use it directly in any browser (no localhost references).

Note: On the Render free tier, cold starts may take 30–60 seconds for the first request after idle time.
The UI includes a small message (“Server is waking up...”) when the /api/health endpoint responds slowly.

Persistence Notes (Render Free Tier)

• SQLite database (database/app.db) is stored on ephemeral disk.
  • Data (users, datasets, history) will reset on redeploy or container restart.
  • The dashboard also displays a notice: “Data resets on server redeploy.”
• For production-ready persistence:
  • Use Render PostgreSQL (free for 90 days), and adapt the DB layer to use PostgreSQL instead of SQLite.

Academic Presentation Notes

• Backend Stack: Python 3.10, Flask, Gunicorn, SQLite.
• Frontend Stack: HTML5, CSS3, vanilla JavaScript, Chart.js.
• Machine Learning:
  • ARIMA via statsmodels
  • LSTM via tensorflow-cpu + keras
  • Hybrid model combining ARIMA and LSTM forecasts.
• Security:
  • Bcrypt password hashing (flask-bcrypt)
  • Server-side session management (flask-session)
  • Per-user access control for datasets and forecast history.