Perbandingan Kinerja Metode Arima dan LSTM Berbasis Machine Learning dalam Meramalkan Cuaca di Pulau Sumatera

Abstract

Weather forecasting is a major component in life that affects various sectors ranging from agriculture to aviation. The research aims to provide information regarding a good method in forecasting weather, especially on the island of Sumatra. This will provide information to certain parties to be able to cope with uncertain weather changes. Rainfall, temperature and humidity are the variable used to forecast the weather using machine learning-based ARIMA and LSTM methods. This research pays attention to the variability that occurs in weather data, such as rainfall data which generally contains outlier data, consistent and stable temperature and humidity data. The ARIMA method uses Autoregressive, Differencing and Moving Average components to create a model to be tested and this method is able to process data that has a linear pattern. The LSTM method uses the epoch model, batch size to create and run the model and this method is able to process data that has non-linear patterns and long-term dependencies. These two methods each provide a match to the weather parameter data. The results show that the ARIMA method provides good forecasting accuracy for humidity data. For rainfall and temperature variable, the LSTM method shows better forecasting accuracy than the ARIMA method. The selection of the right method depends on the characteristics of the weather data so it is necessary for further analysis to get accurate predictions.