Penerapan Data Mining untuk Prediksi Jumlah Produksi Minyak Sawit dengan Metode Long Short-Term Memory

Abstract

Palm oil production is one of the important indicators in the agribusiness sector in Indonesia. Accurate prediction of the amount of palm oil production can help in better planning and decision-making processes. The problem faced is the fluctuation in the number of palm fruit bunches entering the mill, which causes uncertainty in the production of crude oil or CPO. The changing results of CPO production have an impact on the company. This research aims to build a palm oil production prediction model using the Long Short-Term Memory (LSTM) method. The data used includes palm oil production from 2019 to 2023, with data up to 2022 used as the training set and data from 2023 and above as the testing set. The data was normalized using MinMaxScaler to ensure the data was within the appropriate range for LSTM model training. The LSTM model consists of four layers, each with 50 memory units, and comes with a 20% Dropout layer to prevent overfitting. The model is compiled using RMSprop optimizer and Root Mean Squared Error (RMSE) loss function. The model training was conducted for 50 epochs with a batch size of 32. The evaluation results show that the LSTM model is able to provide palm oil production prediction with RMSE values of 0.1238 for total received Palm Fruit Bunches, 0.1177 for total processed Palm Fruit Bunches, and 0.1177 for total CPO. These results show that the LSTM model has good potential in predicting palm oil production, but still requires further improvement.