Employing long short-term memory and Facebook prophet model in air temperature forecasting

• Published in: Communications in statistics. Simulation and computation Vol. 52; no. 2; pp. 279 - 290
• Main Authors: Toharudin, Toni, Pontoh, Resa Septiani, Caraka, Rezzy Eko, Zahroh, Solichatus, Lee, Youngjo, Chen, Rung Ching
• Format: Journal Article
• Language: English
• Published: Philadelphia Taylor & Francis 01.02.2023; Taylor & Francis Ltd
• Subjects: Air temperature; Big Data; Forecasting; LSTM; Machine learning; Mathematical models; Missing data; Neural networks; Outliers (statistics); Prophet model; Short term; Social networks; Temperature; Weather forecasting
• ISSN: 0361-0918; 1532-4141
• DOI: 10.1080/03610918.2020.1854302

One of information needed in weather forecast is air temperature. This value might change any time. Prediction of air temperature is very valuable for some communities and occasions. Therefore, high accuracy prediction is needed. Since the information about air temperature might vary over time, it is necessary to implement methods that can adapt to this situation. The use of neural network methods such as long short term memory (LSTM), nowadays, becomes popular in facing big data including unexpected fluctuation on the data. Thus, the model is used in this paper which provides long series data on air temperature. In addition, recently, Facebook announced an accurate method of forecasting, called Prophet model's, for data which have trend, seasonality, holidays, missing data, not to mention outliers. Hence, the forecast of five-year daily air temperatures in Bandung on this paper is modeled by LSTM and Facebook Prophet. The result shows that, for minimum temperature, Prophet performs better on maximum air temperature while LSTM performs better on minimum air temperature. However, the difference on the value of RMSE is not too large significant.