Please use this identifier to cite or link to this item: http://repository.ipb.ac.id/handle/123456789/168991
Title: Prediksi Hasil Produksi Kebun Kelapa Sawit Melalui Pendekatan Time Series dengan Metode Recurrent Neural Network
Other Titles: 
Authors: Pertiwi, Setyo
Rifki, Muhamad Fahri Nur
Issue Date: 2025
Publisher: IPB University
Abstract: Produksi kelapa sawit yang optimal menjadi kunci keberhasilan perusahaan perkebunan dalam mencapai target produksi. Namun, ketidakakuratan dalam memprediksi hasil produksi dapat menyebabkan penggunaan sumber daya yang kurang efisien. Penelitian ini bertujuan untuk mengembangkan model prediksi hasil produksi kelapa sawit menggunakan pendekatan multivariate time series dengan metode recurrent neural network (RNN). Data yang digunakan meliputi parameter pendukung produksi seperti umur tanaman, luas lahan, jumlah pokok produktif, temperatur, curah hujan, dan lama penyinaran matahari pada tahun 2015-2024. Model dibangun dengan inisialisasi hyperparameter menggunakan arsitektur terbaik (optimizer Adam, learning rate 0,001, batch size 16, epoch 150) dan diuji dengan pendekatan time steps. Hasil evaluasi menunjukkan bahwa model dengan time steps 12 menghasilkan nilai mean absolute percentage error (MAPE) sebesar 5,58% dan root mean squared error (RMSE) sebesar 258,53 yang termasuk kategori prediksi sangat baik. Model yang dikembangkan kemudian diimplementasikan ke dalam website, memungkinkan pengguna dapat mengunggah data historis, mengelola data pada database, dan memperoleh prediksi hasil produksi pada periode mendatang. Penelitian ini menyimpulkan bahwa model RNN efektif dalam mengenali pola produksi jangka panjang dan dapat diterapkan sebagai sistem prediksi untuk mendukung perencanaan serta pengambilan keputusan di perusahaan perkebunan.
Optimal palm oil production is the key to the success of plantation companies in achieving production targets. However, inaccuracies in predicting production yields could lead to inefficient use of resources. This study aimed to develop a palm oil production prediction model using a multivariate time series approach with the recurrent neural network (RNN) method. The data used included supporting production parameters such as plant age, land area, number of productive trees, temperature, rainfall, and sunlight duration from 2015-2024. The model was built by initializing hyperparameters using the best architecture (Adam optimizer, learning rate of 0.001, batch size of 16, and 150 epochs) and tested using a time steps approach. Evaluation results showed that the model with 12 time steps produced a mean absolute percentage error (MAPE) of 5.58% and a root mean squared error (RMSE) of 258.53, which was categorized as very good prediction. The developed model was then implemented into a website, enabling users to upload historical data, manage the database, and obtain production predictions for upcoming periods. This study concluded that the RNN model was effective in recognizing long-term production patterns and could be applied as a prediction system to support planning and decision-making in plantation companies.
URI: http://repository.ipb.ac.id/handle/123456789/168991
Appears in Collections:UT - Agricultural and Biosystem Engineering

Files in This Item:
File Description SizeFormat 
cover_F1401211089_04ec6971bf1145d68cbe5c9bba24e8d8.pdfCover2.11 MBAdobe PDFView/Open
fulltext_F1401211089_7ed91d2a1d29449bba01f6b543770e5a.pdf
  Restricted Access
Fulltext9.3 MBAdobe PDFView/Open
lampiran_F1401211089_29ad9aaa655248cf8aa12706da2441db.pdf
  Restricted Access
Lampiran2.72 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.