Prophet Model Performance Analysis for Jakarta Air Quality Forecasting

The current condition of air quality requires special attention, especially in the city of Jakarta. Based on the Air Quality Index (AQI) website, Jakarta is in sixth place which the city with the worst air quality in the world an AQI value of 156 (unhealthy). One of the things that can overcome this problem is forecasting. Forecasting can be useful for the government and society to take the necessary steps towards air quality control in the future. One of the methods for forecasting is Prophet model. In its application, the Prophet model can be implemented with several methods, namely applying holiday components, seasonal additional monthly periods, and hyper-parameter tuning with Bayesian Optimization. The testing process is carried out using daily data from ISPU DKI Jakarta (2010 to 2019) obtained from the Open Data Jakarta website with data from 2010 to 2018 as training data and 2019 data as test data. The results of these tests show that the Prophet model is quite well implemented for forecasting most of the pollutant parameters, namely PM10, SO2, CO, and O3. Then, the application of holiday components, additional seasonal monthly periods, and hyperparameter tuning with Bayesian Optimization is able to increase the accuracy of the Prophet model on some pollutant parameters. And when compared to ARIMA, the accuracy of the Prophet model is superior to forecasting parameters SO2, CO, and O3, while ARIMA is superior to forecasting PM10 and NO2 parameters.