trained a machine learning model on 45000 data points that had hourly measurements of weather data such as dew, snow, rain, wind speed, wind direction, pressure and pollution. The model predicts pollution based on previous 11 datapoints. Currently model provides satisfactory performance with RMSE score of 0.064.
The user interface uses LSTM model with a look-back window of 11. Other models are also trained such as ARIMA, TCN(temporal convolution networks). Notably TCN provides best performance.
