Machine Learning-Based Air Quality Index Prediction System: From Data to Interactive Dashboard

This article introduces an end-to-end machine learning project that achieves accurate Air Quality Index (AQI) prediction, including comparisons of 11 algorithms, over 20 feature engineering techniques, and a Streamlit interactive dashboard. The project achieves a prediction accuracy of 95.14% and covers the entire workflow from data collection, feature engineering, model training to visualization, providing a complete reference for environmental data science applications.

Project Objective: Predict the comprehensive AQI using sensor data such as PM2.5, PM10, NO₂, CO, temperature, and humidity. The tech stack includes: Scikit-learn (machine learning framework), Streamlit (interactive dashboard), Plotly/Matplotlib/Seaborn (visualization), Pandas/NumPy (data processing), balancing practicality and advancement.

Data Cleaning: Automatically handle missing values and outliers; Feature Engineering Innovations: Construct over 20 derived features from 6 original parameters (e.g., composite pollution index pm_avg=(PM2.5+PM10)/2, interaction feature pollution_humidity=PM2.5×Humidity, nonlinear transformation pm25_squared, etc.); Feature Selection: Retain a subset of features with high predictive power through correlation analysis and redundancy detection.

Compare 11 algorithms (linear, tree, ensemble, etc.), use 5-fold cross-validation (R²=0.9222±0.0134) to ensure stability, and optimize hyperparameters with RandomizedSearchCV. The final Lasso regression performed best: test set R²=0.9514, MAE=3.68 AQI points, RMSE=4.61.

The Streamlit dashboard includes 5 modules: Home Overview (key indicators), Data Explorer (correlation heatmap, etc.), Real-time Prediction (input parameters to get AQI and health recommendations), Model Performance Analysis, and Methodology Documentation. Built-in EPA standard AQI classification (0-50 Good, 51-100 Moderate, etc.).

Engineering Highlights: Avoid data leakage, modular code structure, support for multi-platform deployment (Streamlit Cloud/Heroku/Docker). Application Scenarios: Smart city early warning, health management recommendations, educational and research cases, industrial monitoring compliance. Expansion Directions: Introduce time series models, integrate satellite data, develop mobile applications.

Key Success Factors of the Project: Solid feature engineering, rigorous model evaluation, user-centric design. Provides a complete project reference for data science learners; its open-source nature supports improvement and innovation, reflecting the value of knowledge sharing in driving technological progress.