Comparison Vector Autoregressive and Long Short Term Memory for forecasting Air Pollution Index in Jakarta

The Air Pollutant Index (API) is a number without units that describes the condition of ambient air quality in a certain location. API monitoring is carried out based on meteorological data affecting ambient air concentrations. On the SILIKA DKI Jakarta website, there is no feature to see ISPU predictions even though this feature is useful for the community. The benefit of predicting ISPU is that people can anticipate early air quality conditions that will occur, including air pollution. In the time series study, a comparison of multivariate time series problems was carried out on ISPU parameters that were correlated with each other. Conventional methods such as Vector Autoregressive. The VAR model is a development of the Autoregressive (AR) model where more than one endogenous variable is used in the VAR model. Artificial Intelligence forecasting methods such as Long Short-Term Memory can be used to forecast multivariate time series. The purpose of this research is to compare the best method between VAR and LSTM in forecasting the parameter index of air pollutant standards in Jakarta. Air pollutant standard index data comes from the Jakarta environmental service from January 2021 to December 2021. These two methods are compared using the lowest RMSE value so that the best method is obtained. The results showed that the LSTM model had the best RMSE value for the prediction of pm10, which was 24.482, and pm25, which was 22.504. Modeling for multivariate time series can be done by conventional methods such as VAR. However, in this study, the deep learning algorithm, namely long short-term memory, is the best method that can be used to solve multivariate time series cases. In addition, the LSTM method is a method that is easier to implement because it does not require parameter estimation like the VAR model.