An end-to-end Air Quality Index (AQI) analysis and forecasting platform. This project integrates a comprehensive Power BI analysis, a robust Flask backend, a dynamic vanilla JavaScript frontend dashboard, and machine learning pipelines. The entire project was developed iteratively in phases, focusing on accurate data representation and predictive analytics.
The foundational dataset comprises historical AQI, weather, demographics, and date-dimension data. To ensure data integrity, some synthetic data was strategically incorporated to fill gaps and maintain consistent coverage.
The data cleaning pipeline was executed in the following steps:
- Negative Value Removal: Extraneous negative (
-ve) values were identified and eliminated. - Precision Rounding: Relevant numerical columns were rounded to two decimal places for consistency.
- Data Thresholding: Cities with less than 5 years (60 months) of historical data were removed to maintain statistical significance.
- Date Dimension Creation: A dedicated Date Dimension was generated to support robust time-series analysis.
- Entity Cascading: Cities that failed to meet the threshold were removed from the main table, and these deletions were strictly cascaded to all associated relational tables.
- Country Updates: Similarly, country references were updated and synchronized across the entire schema.
The underlying architecture utilizes a star/snowflake schema linking AQI data, cities, countries, demographics, weather, and a date dimension.
Following the data cleaning, an initial analysis was conducted using Power BI to establish a strong analytical baseline. This phase explored historical trends, pollutant distributions, and geographical comparisons.
Note: Each Power BI dashboard view is equipped with advanced drill-through features, enabling deep dives into specific cities, years, and pollutant metrics.
An overarching view of AQI, weather, and demographic contexts.
Detailed breakdown of individual pollutant metrics (AQI, CO, NH3, NO2, O3, PM2.5, PM10, SO2) over time.
Analysis of AQI category counts (Good, Moderate, Unhealthy, etc.) and pollutant distribution outliers.
The project leverages three core modeling approaches to derive insights and forecast future trends. The resulting model artifacts are serialized and saved using joblib (as .pkl files) for efficient instantiation in the production backend.
-
Random Forest Regressor
- Purpose: Feature importance extraction.
- Details: Analyzes historical data to determine the most significant pollutants driving the overall AQI.
-
XGBoost (Extreme Gradient Boosting)
- Purpose: Time-series AQI forecasting.
- Details: Predicts future monthly AQI values by evaluating historical lags, weather conditions, and demographic features.
-
Statistical Correlation Model
- Purpose: Pollutant influence mapping.
- Details: Computes city-level correlation matrices to map the relationships between different environmental factors (e.g., how humidity correlates with PM2.5).
Following the data preparation, Power BI analysis, and machine learning modeling, a dedicated web platform was developed to serve the insights interactively.
Users can explore historical AQI behavior and identify dominant pollutant contributors across different cities on an interactive global map.
The dashboard allows users to select a specific city and visualize historical AQI trends over custom date ranges.
Identify the highest contributing pollutants to the AQI and observe how environmental factors influence pollutant levels.
View the demographic context for each city (urbanization, population, industrial area) alongside XGBoost-powered future monthly AQI forecasts.
AQI/
├── README.md
├── requirements.txt
├── artifacts/ # Generated JSON/CSV outputs (correlations, feature importance)
├── backend/ # Flask API source code
├── data/ # Cleaned datasets
├── frontend/ # Vanilla JS/HTML/CSS dashboard
├── models/ # joblib (.pkl) model artifacts
├── notebooks/ # ML training and correlation notebooks
├── Screenshots/ # Dashboard visualizations
└── Power_BI_dashboard.pbix # Original Power BI dashboard file
Note: env/, .git/, and __pycache__/ are local workspace/runtime folders and are not part of the documented project source tree.
-
Install Dependencies
pip install -r requirements.txt
-
Generate Models (Optional) Run the notebooks in
notebooks/sequentially to retrain the models and refresh the generated outputs inartifacts/andmodels/. -
Start the Backend Server
python backend/app.py
-
Launch the Application Open the application in your browser at
http://127.0.0.1:5000/.
The Flask backend exposes several endpoints for data retrieval and prediction:
# Retrieve city-specific AQI data
curl "http://127.0.0.1:5000/get_city_data?city_name=Delhi"
# Retrieve global feature importance scores
curl "http://127.0.0.1:5000/get_feature_importance"
# Retrieve pollutant correlations for a city
curl "http://127.0.0.1:5000/get_pollutant_correlation?city_name=Delhi"
# Generate future AQI predictions using the XGBoost model
curl "http://127.0.0.1:5000/predict_aqi?city_name=Delhi&horizon=12"- Coordinates: The dataset does not include latitude and longitude; the map utilizes deterministic estimated coordinates for city placement.
- Synthetics: Some demographic and weather values are synthetic, used to ensure uninterrupted coverage across the timeline.
- Error Handling: If
joblibmodel files or artifact files are missing from themodels/orartifacts/directories, the backend API will return a descriptive error indicating which notebook must be run first.