AI-Driven Air Quality Prediction System: Fusion of ML and Generative AI

This is an end-to-end AI project aligned with UN Sustainable Development Goal 13 (Climate Action). It combines Random Forest regression and Google Gemini large model to achieve dual functions: air quality index (AQI) prediction and health advice generation. The system stands out with its 'prediction + explanation' capability, serving as an intelligent environmental assistant for end users.

Project Background

Air pollution is a global environmental health challenge. According to WHO, over 90% of the world's population breathes air that does not meet safety standards, leading to millions of premature deaths annually. Predicting AQI accurately and providing understandable health advice is of great social value.

The Air-Quality-Prediction-Model project targets SDG13. It integrates traditional machine learning and cutting-edge generative AI, not only predicting AQI based on pollutant data but also generating human-readable pollution explanations and health recommendations, distinguishing itself from pure numerical prediction tools.

System Architecture & Tech Stack

The system uses modular design:

Data Modeling Layer: Core is a Random Forest regression model. Chosen for its efficiency on structured data (lower computation cost, less tuning) compared to deep learning. Inputs: CO2, NO2, SO2 concentrations; output: AQI.

Generative Explanation Layer: Integrates Google Gemini to generate contextual natural language explanations based on predicted AQI, including pollution level descriptions, cause analysis, and targeted health advice (e.g., outdoor activity suitability, protection measures for sensitive groups).

Service Interface Layer: Flask REST API with two endpoints: /predict (get AQI prediction) and /explain (prediction + AI-generated explanation). Enables integration into smart city platforms.

Deployment Layer: Containerized with Dockerfile and run.sh script for consistent deployment across environments.

Core Functions

Air Quality Prediction: Receives pollutant data, computes AQI via pre-trained Random Forest model. model.py includes data generation logic for quick startup; model is persisted as pickle and loaded on API startup.

Smart Health Advice: The /explain endpoint first gets AQI prediction, then sends a well-designed prompt to Gemini to generate personalized advice. This generative approach is more flexible and expressive than fixed templates.

API Usage: Simple JSON request (with CO2, NO2, SO2) to /explain returns AQI and AI explanation, easy to integrate with mobile apps, web frontends, or IoT devices.

Engineering Practice Highlights

SEAI Integration: Demonstrates Software Engineering in AI (SEAI) best practices—packaging ML models into production-ready services with environment management, dependency isolation, and containerization. A great reference for turning AI prototypes into products.

Dual Model Balance: Random Forest handles deterministic, interpretable numerical prediction; Gemini handles creative, expressive text generation. Both complement each other for a cost-effective, full-featured solution.

Developer Experience: run.sh simplifies Docker build/run; virtual environment config reduces entry barriers, reflecting attention to user experience.

Application Scenarios & Social Value

Personal Health: Helps sensitive groups (asthma patients, elderly, children) adjust outdoor plans to reduce health risks.

City Planning: Assists managers in monitoring air quality trends, identifying pollution hotspots, and evaluating policy effectiveness for data-driven decisions.

Environmental Education: AI-generated explanations make environmental knowledge accessible, raising public awareness and promoting sustainable lifestyles.

Expansion & Improvement Directions

• Data Sources: Integrate real-time monitoring station data or satellite remote sensing data to enhance timeliness and coverage.
• Model Upgrades: Try advanced time-series methods (LSTM, Transformer) to capture dynamic air quality changes.
• Multi-modal Explanations: Add visualizations (pollution maps, trend charts) for richer information presentation.

Conclusion

The Air-Quality-Prediction-Model shows AI's potential to solve real social problems. It is a practical system focused on engineering implementation, balancing performance and cost—not a tech demo. In the face of climate change and pollution, such projects combining technical innovation and social responsibility demonstrate the path of 'tech for good'.