Implementasi Time Series Forecasting dengan Algoritma LSTM untuk Pemantauan dan Prediksi Produktivitas Kelapa Sawit Berdasarkan Hasil Panen

Abstract

Palm oil productivity is a key factor in maintaining the stability and sustainability of Indonesia's agribusiness industry. The fluctuation in yield at PTPN IV Kebun Bah Birung Ulu, which increased from 43,308 tons in 2020 to 44,028 tons in 2022 and then decreased to 34,643 tons in 2024, highlights the need for a more accurate monitoring system. These fluctuations are influenced by weather, fertilizer usage, plant infections, and plant age. Manual record-keeping without digital system support also increases the risk of errors and complicates production monitoring. This study aims to develop a web-based palm oil productivity prediction system using the Long Short-Term Memory (LSTM) algorithm. Five years of daily historical data, including plant age, fertilizer usage, rainfall, infection rates, and harvest results per afdeling, were used as model input. The research process includes data collection, preprocessing with Min-Max normalization, data splitting into 80% training and 20% testing, and training the LSTM model with two LSTM layers, two dropout layers, and one Dense layer. Model evaluation using Root Mean Squared Error (RMSE) shows that the model can predict productivity with good accuracy, with the best RMSE for each target variable achieved at different epochs. The 2025 prediction results indicate a stable or declining trend influenced by plant age, fertilizer application, rainfall, and infection rates. The developed web-based system features real-time monitoring and data visualization, providing a more efficient solution for production management and strategic decision-making in palm oil plantations.