Short-Term AQI Forecasts using Machine/Deep Learning Models for San Francisco, CA

The city of San Francisco, CA is highly susceptible to severe air pollutants and often experiences a poor Air Quality Index (AQI). Some primary pollutants include Carbon Dioxide (CO2), Carbon Monoxide (CO), Nitrogen Oxide (NOx), Particulate Matter (PM), and Sulphur Dioxide (SO2). This paper estimates short-term AQI indices for the city of San Francisco, CA. Ten years of historical AQI datasets were explored for trends, levels, cyclicity, and seasonality to predict for the next 7-day and 30-day window periods. Multiple Machine/Deep Learning models such as Random Forest (RF), Support Vector Regression (SVR), XGBoost (XGB), Neural Network (NN), and Long Short-Term Memory (LSTM) were deployed. The performance of these models are assessed using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) metrics. The preliminary results indicate that the XGBoost outperforms over other models with MAE scores of 7.991 (7-day) and 8.126 (30-day), respectively. We also remind readers that for forecasting real-time AQIs, multiple factors such as smoke pollutants from wildfires, toxic spills from train derailments, and distributed energy resources (DERs) contributors such as electric vehicle, wind, and solar fleets must be taken into consideration for robust accuracy.