Lahore Smart Air Quality Monitoring and Prediction System: Machine Learning-Driven Environmental Data Application

This project is developed by emanjum, with source code available on GitHub (link: https://github.com/emanjum/lahore-aqi-project, release date: June 7, 2026). The core is to build an intelligent system based on OpenWeather API, MongoDB Atlas, multiple machine learning models, and Streamlit, enabling the monitoring and prediction of Lahore's air quality. It covers the entire process from data collection, storage, modeling to visualization, and practices MLOps automation through GitHub Actions.

Lahore is the second-largest city in Pakistan and one of the most air-polluted cities in the world. Its PM2.5 concentration has long exceeded the WHO safety standards, seriously threatening residents' health. This project integrates multiple technology stacks (OpenWeather API, MongoDB Atlas, ML algorithms, Streamlit) to build a complete solution, while using GitHub Actions to implement automated workflows, reflecting modern MLOps best practices and having important practical significance.

The system is divided into four modules:

1. Data Collection Layer: Obtains real-time AQI, PM2.5 and other pollutant indicators and meteorological data through the OpenWeather API, with regular collection to ensure timeliness.
2. Data Storage Layer: Uses MongoDB Atlas for storage, thanks to its flexible document model, horizontal scalability, and high availability. Data is organized in time series for easy analysis.
3. Model Training Layer: Adopts three models: linear regression (baseline), random forest (non-linear), and XGBoost (advanced integrated solution).
4. Visualization Layer: Builds an interactive dashboard using Streamlit, displaying real-time data, historical trends, prediction results, and model performance comparisons.

The project uses three models:

• Linear Regression: A basic supervised learning method that assumes a linear relationship. Its advantages are simplicity and interpretability, while its disadvantage is difficulty in capturing non-linear relationships.
• Random Forest: An ensemble of multiple decision trees. It handles non-linear interactions, is robust to outliers, and can discover the impact of complex combinations of factors.
• XGBoost: An advanced implementation of gradient-boosted decision trees. It iteratively corrects errors and uses regularization to prevent overfitting, achieving high prediction accuracy and excellent performance in competitions.

The project uses GitHub Actions to implement automated workflows:

• Scheduled Data Collection: Regularly pulls OpenWeather data to update the database without manual intervention.
• Model Retraining: Automatically triggered after accumulating new data to solve the problem of model drift.
• Automated Testing: Runs unit/integration tests when code is submitted to ensure quality.
• Continuous Deployment: Trained models are automatically deployed to the production environment to update prediction services. This workflow reduces operational burden and allows focus on model optimization.

The system's value is reflected in multiple aspects:

• Public Health: Residents can check real-time AQI and predictions to guide decisions on going out and protection, especially benefiting sensitive groups.
• Government Decision-Making: Environmental protection departments formulate pollution control measures (such as traffic restrictions, production suspension, etc.) based on the data.
• Research Accumulation: Long-term data is used for academic research such as climate change and pollution source analysis.
• Commercial Applications: Air purifier manufacturers adjust their marketing strategies, and insurance companies develop health insurance products.

Possible expansion directions for the project:

• Multi-site Coverage: Expand from the main urban area to the entire city or multiple cities nationwide.
• Diversified Data Sources: Integrate local monitoring stations, low-cost sensors, and satellite data to improve accuracy and coverage.
• Model Upgrades: Try deep learning models such as LSTM and Transformer to capture complex time dependencies.
• Early Warning System: Automatically send warnings when AQI exceeds the threshold.
• Mobile Application: Develop a mobile App for the public to check anytime.

This project is a complete data science example, covering the entire process from data collection to deployment. It demonstrates the ability of modern technology stacks to solve environmental problems and reflects the value of machine learning in public services. It serves as an end-to-end project reference for beginners and provides an extensible technical framework for researchers.