Peramalan Produksi Kelapa Sawit dengan Pendekatan Hierachical Time Series Extreme Gradient Boosting dan Neural Network Autoregressive

Mahmudah, Nazia (2025) Peramalan Produksi Kelapa Sawit dengan Pendekatan Hierachical Time Series Extreme Gradient Boosting dan Neural Network Autoregressive. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003211157-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Perubahan pola curah hujan yang tidak menentu berdampak signifikan terhadap produksi kelapa sawit. Penelitian ini bertujuan untuk meramalkan produksi kelapa sawit tahun 2025 secara akurat dengan mempertimbangkan struktur hierarki wilayah dan dinamika musiman curah hujan. Metode yang digunakan adalah Extreme Gradient Boosting (XGBoost) dan Neural Network Autoregressive (NNAR) dengan variabel eksogen curah hujan serta pendekatan Hierarchical Time Series (HTS). Data yang digunakan meliputi curah hujan dan yield dari 9 area dan 3 wilayah perkebunan pada periode 2016–2024. Model dievaluasi menggunakan metrik MAPE, RMSE, dan kemiripan pola melalui pendekatan rolling window 4 tahunan. Hasil evaluasi menunjukkan bahwa XGBoost memiliki performa lebih baik dibandingkan NNAR, dengan nilai rata-rata MAPE sebesar 18,38% dan RMSE 0,29, sedangkan NNAR mencatat MAPE 24,51% dan RMSE 0,37. Dalam implementasi HTS, metode Bottom-Up menunjukkan akurasi terbaik secara keseluruhan pada tingkat perusahaan (MAPE 13,49%; RMSE 14.675,13 ton) dan area (MAPE terendah 13,08% di Area 1; RMSE terendah 400,24 di Area 9). Sementara itu, metode Top-Down lebih unggul di tingkat wilayah, khususnya Sumatera (MAPE 11,32%; RMSE 400,24). Hasil peramalan 2025 menunjukkan pola produksi yang stabil dan realistis setelah diagregasi. Temuan ini memberikan implikasi strategis dalam pengambilan keputusan berbasis data, terutama untuk perencanaan panen, logistik, dan mitigasi risiko iklim di industri kelapa sawit. ==========================================================================================================================================
The unpredictable changes in rainfall patterns significantly impact palm oil production. This study aims to accurately forecast palm oil production in 2025 by incorporating the hierarchical structure of plantation regions and seasonal rainfall dynamics. The methods employed are Extreme Gradient Boosting (XGBoost) and Neural Network Autoregressive (NNAR), with rainfall as an exogenous variable, using a Hierarchical Time Series (HTS) approach. The dataset comprises rainfall and yield data from 9 areas across 3 plantation regions for the period 2016–2024. Models were evaluated using MAPE, RMSE, and pattern similarity through a 4-year rolling window approach. Evaluation results show that XGBoost outperformed NNAR, achieving an average MAPE of 18.38% and RMSE of 0.29, whereas NNAR recorded a MAPE of 24.51% and RMSE of 0.37. In the HTS implementation, the Bottom-Up method demonstrated the best overall accuracy at the company level (MAPE 13.49%; RMSE 14,675.13 tons) and area level (lowest MAPE of 13.08% in Area 1; lowest RMSE of 400.24 in Area 9). Meanwhile, the Top-Down method performed better at the regional level, particularly in Sumatra (MAPE 11.32%; RMSE 400.24). The 2025 forecast results indicate a stable and realistic production pattern after aggregation. These findings offer strategic implications for data-driven decision-making, especially in harvest planning, logistics, and climate risk mitigation within the palm oil industry.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Curah Hujan, Hierarki Deret Waktu, Kelapa Sawit, NNAR, Peramalan Produksi, Rolling Window Forecasting, XGBoost, Hierarchical Time Series, NNAR, Palm Oil, Production Forecasting, Rainfall, Rolling Window Forecasting, XGBoost
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis Q Science > Q Science (General) Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. S Agriculture > S Agriculture (General) S Agriculture > S Agriculture (General) > S600.7.R35 Rain and rainfall
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Nazia Mahmudah
Date Deposited:	31 Jul 2025 06:37
Last Modified:	31 Jul 2025 06:37
URI:	http://repository.its.ac.id/id/eprint/124196

Actions (login required)

View Item