Air Temperature Forecasting with Long Short-Term Memory and Prophet: A Case Study of Jakarta, Indonesia
Mohammad Daffa Haris, Didit Adytia, Annas Wahyu Ramadhan
Abstract
The high number of industrial and residential areas has reduced green space in Jakarta. This condition increases air temperature, contributing to climate change in Jakarta and most other big cities in Indonesia. Therefore, an accurate air temperature prediction model is needed to support daily public activities. On the other hand, the government can also use this prediction to determine regulations to suppress climate change. This study developed Jakarta’s air temperature prediction model using two machine learning models: Long Short-Term Memory (LSTM) and Prophet. LSTM is a variant of the classic Recurrent Neural Networks (RNN) with the addition of memory blocks that stores long-term information. The Prophet is an additive regression model developed by Facebook. These models are chosen to handle stochastic data such as air temperature. Here, we forecast the time series of air temperature based on sequential historical data. The accuracy of prediction is measured by using RMSE and Correlation Coefficient values. Results of the study indicate that the LSTM performs better for short-term forecasts, i.e., 2 to 48 hours, with RMSE values between 0.31 to 0.69. On the other hand, the Prophet model is suitable for more long-term predictions, i.e., 72 to 168 hours, with RMSE between 0.80 and 0.89.