Perancangan Sistem Prediktor Cerdas untuk Prediksi Hasil Panen Buah Sawit Berbasis Machine Learning

Darmawan, Tiffany (2026) Perancangan Sistem Prediktor Cerdas untuk Prediksi Hasil Panen Buah Sawit Berbasis Machine Learning. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6009241007-Master_Thesis.pdf Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Indonesia merupakan negara produsen kelapa sawit terbesar di dunia sehingga akurasi prediksi produksi menjadi salah satu faktor penting dalam mendukung perencanaan industri, pengelolaan sumber daya, dan stabilitas rantai pasok. Produksi kelapa sawit dipengaruhi oleh berbagai faktor cuaca yang memiliki hubungan nonlinier dan menunjukkan keterlambatan respons biologis tanaman terhadap perubahan kondisi lingkungan. Penelitian ini bertujuan untuk merancang sistem prediktor cerdas berbasis machine learning, mengevaluasi kinerja beberapa model prediksi hasil panen kelapa sawit, dan menganalisis pengaruh variabel cuaca terhadap hasil prediksi. Dataset yang digunakan berupa data bulanan produksi kelapa sawit dan data cuaca periode 2017–2025 di Kabupaten Pasangkayu, Sulawesi Barat. Variabel cuaca yang digunakan meliputi curah hujan, jumlah hari hujan, temperatur, kelembapan udara, kecepatan angin, dan lama penyinaran matahari. Penelitian menerapkan pendekatan lag-time 6 bulan, 12 bulan, dan 24 bulan untuk merepresentasikan respons biologis tanaman kelapa sawit terhadap perubahan kondisi cuaca. Model yang dikembangkan terdiri atas Artificial Neural Network (ANN), Random Forest, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), dan Support Vector Regression (SVR). Hasil penelitian menunjukkan bahwa model XGBoost dengan lag-time 12 bulan memberikan performa terbaik dibandingkan model lainnya dengan nilai RMSE sebesar 323,31 ton, MAE sebesar 99,60 ton, MAPE sebesar 2,36%, dan R² sebesar 0,89. Hasil menunjukkan bahwa model mampu menjelaskan sekitar 89% variasi data produksi kelapa sawit dengan tingkat kesalahan prediksi yang rendah. Analisis feature importance menunjukkan bahwa lama penyinaran matahari (Sunshine) merupakan variabel cuaca yang paling berpengaruh terhadap prediksi hasil panen kelapa sawit dengan nilai importance sebesar 15,25%, jumlah hari hujan (Rainy Days) sebesar 9,07%, temperatur sebesar 7,14%, curah hujan sebesar 4,08%, kelembapan udara sebesar 3,35%, dan kecepatan angin sebesar 3,10%.
=================================================================================================================================
Indonesia is the world's largest producer of palm oil, making production forecasting accuracy an important factor in supporting industrial planning, resource management, and supply chain stability. Palm oil production is influenced by various weather-related factors that exhibit nonlinear relationships and delayed biological responses to environmental changes. This study aims to develop an intelligent prediction system based on machine learning, evaluate the performance of several prediction models for palm oil yield forecasting, and analyze the influence of weather variabels on prediction results. The dataset used in this study consists of monthly palm oil production and weather data collected from 2017 to 2025 in Pasangkayu Regency, West Sulawesi, Indonesia. The weather variabels include rainfall, rainy days, temperature, humidity, wind speed, and sunshine duration. Lag-time approaches of 6, 12, and 24 months were applied to represent the biological response of oil palm plants to weather conditions. The developed models include Artificial Neural Network (ANN), Random Forest, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Support Vector Regression (SVR). The results show that the XGBoost model with a 12-month lag-time achieved the best performance among all evaluated models, with an RMSE of 323.31 tons, MAE of 99.60 tons, MAPE of 2.36%, and an R² value of 0.89. These results indicate that the model can explain approximately 89% of the variation in palm oil production while maintaining a low prediction error. Feature importance analysis revealed that sunshine duration was the most influential weather variabel affecting palm oil yield prediction, with an importance value of 15.25%, followed by rainy days (9.07%), temperature (7.14%), rainfall (4.08%), humidity (3.35%), and wind speed (3.10%). This study demonstrates that the XGBoost-based machine learning approach can be effectively utilized for palm oil yield prediction using weather data.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Kelapa Sawit, Machine Learning, Variabel Cuaca, Sistem Prediktor, Supervised Learning ==================== Oil Palm, Machine Learning, Weather Variabel, Predictor System, Supervised Learning
Subjects:	S Agriculture > S Agriculture (General) > S600.7.P53 Planting time S Agriculture > S Agriculture (General) S Agriculture > S Agriculture (General) > S600.7.R35 Rain and rainfall T Technology > T Technology (General) T Technology > T Technology (General) > T58.8 Productivity. Efficiency
Divisions:	Faculty of Industrial Technology > Physics Engineering > 30101-(S2) Master Thesis
Depositing User:	Tiffany Rachmania Darmawan
Date Deposited:	31 Jul 2026 02:01
Last Modified:	31 Jul 2026 02:01
URI:	http://repository.its.ac.id/id/eprint/140486

Actions (login required)

View Item