Delhi AQI Forecasting System

A production-style Streamlit forecasting system for Delhi Air Quality Index (AQI/PM2.5).

System Architecture

+-------------------+       +----------------------+       +-----------------------+
|   OpenAQ API      | ----> |   b) Offline Update  | ----> |   Historical Data     |
| (Historical Data) |       | (fetch_history.py)   |       | (Standardized CSV/DB) |
+-------------------+       +----------------------+       +-----------+-----------+
                                                                       |
                                                                       v
                                                           +-----------------------+
                                                           |   Feature Engineering |
                                                           |   (make_features.py)  |
                                                           +-----------+-----------+
                                                                       |
+-------------------+       +----------------------+                   v
|   OpenAQ API      | ----> |   Inference Engine   |       +-----------------------+
| (Real-time Data)  |       | (fetch_recent.py)    |       |   Model Training      |
+-------------------+       | (predict.py)         | <---- |   (train_models.py)   |
          |                 +-----------+----------+       |   (XGBoost 6h/12h/24h)|
          |                             |                  +-----------------------+
          v                             v
+--------------------------------------------------+
|               Streamlit Dashboard                |
|           (User selects 6h / 12h / 24h)          |
+--------------------------------------------------+

Directory Structure

• /data: Scripts for data acquisition and storage.
• /training: Offline training pipeline (cleaning, feature engineering, modeling).
• /inference: Online inference logic (fetching recent data, generating predictions).
• /models: Trained model artifacts (.pkl).
• app.py: Main Streamlit application.

Quick Start

1. Setup

pip install -r requirements.txt

2. Data Acquisition (Offline)

Fetch historical data (last 1 year +):

python data/fetch_history.py

3. Training

Train the XGBoost models for 6h, 12h, and 24h horizons:

# 1. Clean and resample to hourly
python training/build_hourly.py

# 2. Generate features and target
python training/make_features.py

# 3. Train models
python training/train_models.py

Models will be saved to the /models directory.

4. Running the App

streamlit run app.py

Deployment

Deploying to Streamlit Cloud (Free & Easy)

1. Push to GitHub:

   • Create a GitHub repository.
   • Push all files, including the /models directory (this is crucial!).
   • Make sure requirements.txt is in the root.

2. Connect to Streamlit Cloud:

   • Go to share.streamlit.io and log in.
   • Click "New App".
   • Select your GitHub repository.
   • Set Main file path to app.py.

3. Add Your API Key (Secrets):

   • Once deployed (or before), go to the app's Settings -> Secrets.
   • Add your OpenAQ key like this:

     OPENAQ_API_KEY = "your_actual_api_key_here"

   • Save. The app will restart and automatically pick up the key!

Running Locally

To run locally, you can create a .env file in the root:

OPENAQ_API_KEY=your_key_here

Or verify it works by creating .streamlit/secrets.toml.