Prediction of Indonesian Palm Oil Production Using Long Short-Term Memory Recurrent Neural Network (LSTM-RNN)

Abstract

At present, the plantation sector is one of the biggest contributors to Indonesia’s Gross Domestic Product (GDP). However, due to fluctuating annual production of Indonesian palm oil, the government is confused in determining palm oil import or export policies. Therefore, a good method is needed to predict Indonesian palm oil production. Soft computing can be used for classification and prediction. Soft computing is a model approach to compute by mimicking the ability of extraordinary human reason to reason and learn in environments that have uncertainty and inaccuracy. Some techniques in soft computing include fuzzy systems, artificial neural networks, evolutionary algorithms, and probabilistic reasoning. One method in Artificial Neural Network is Recurrent Neural Network (RNN). RNN is that the network contains at least one feed-back connection, so the activations can flow round in a loop. In the last few years, the RNN network model has been developed, namely by using the Long Short-Term Memory (LSTM) layer. By using the LSTM layer, the RNN learning process gets better. Therefore, in this study the prediction of palm oil production using the LSTM-RNN method is based on the time series data from 1970 to 2017. The results of this study are found that the LSTM-RNN method is very well used for predictions because it produces MAPE of 2.7098% for training data and 2.9861% for testing data compared to other prediction methods.