Please use this identifier to cite or link to this item:
http://repository.ipb.ac.id/handle/123456789/154592
Title: | Komparasi Kinerja Algoritma XGBoost, CatBoost, dan LightGBM dalam Klasifikasi Produktivitas Sawit Rakyat di Provinsi Riau |
Other Titles: | Peformance Comparison of XGBoost, CatBoost, and LightGBM Algorithms in Classifying Smallholder Palm Oil Productivity in Riau Province |
Authors: | Afendi, Farit Mochamad Sulvianti, Itasia Dina Arss, Muhammad Nachnoer Novatron Fitra |
Issue Date: | 2024 |
Publisher: | IPB University |
Abstract: | Perkebunan sawit di Provinsi Riau memegang peran sentral dalam industri
perkebunan nasional. Faktor utama yang memengaruhi tingkat produktivitas sawit
yaitu perbedaan karakteristik iklim tiap wilayah dan variabilitas iklim. Studi ini
bertujuan membandingkan algoritma machine learning yaitu XGBoost, CatBoost,
dan LightGBM yang digunakan dalam pemodelan iklim untuk mengklasifikasikan
tingkat produktivitas sawit rakyat di Provinsi Riau. Data yang digunakan
merupakan data Statistik Perkebunan Indonesia 1990-2022 dan data iklim yang
didapatkan dari Climate Hazards Group Precipitation with Station (CHIRPS).
Pemodelan dievaluasi melalui tiga skenario partisi model terbaik hasil
hyperparameter tuning. Hasil penelitian model klasifikasi produktivitas sawit
rakyat di Provinsi Riau menunjukkan bahwa CatBoost memiliki AUC uji terbaik
yaitu sebesar 0,857 pada skenario partisi 75% data latih dan 25% data uji.
Kombinasi hyperparameter yang digunakan yaitu ntree = 100, learning rate = 0,03,
dan max-depth = 9. CatBoost juga menunjukkan konsistensi terbaik dalam
menghindari overfitting untuk tiap skenario yang diuji. LightGBM dan XGBoost
memiliki performa yang lebih rendah karena cukup sensitif terhadap data
berdimensi kecil dalam pelatihan. Hal ini menyebabkan inkonsistensi kinerja
algoritma ketika diterapkan pada proporsi data latih dan data uji yang berbeda. Oil palm plantations in Riau Province play a central role in the national plantation industry. The main factors affecting oil palm productivity are the differences in climate characteristics across regions and climate variability. This study aims to compare the machine learning algorithms XGBoost, CatBoost, and LightGBM, used in climate modeling to classify the productivity levels of smallholder oil palm plantations in Riau Province. The data used includes Indonesian Plantation Statistics from 1990 to 2022 and climate data from the Climate Hazards Group Precipitation with Station (CHIRPS). The modeling was evaluated through three partitioning scenarios of the best models resulting from hyperparameter tuning. The results showed that CatBoost had the best test AUC of 0.857 in the scenario with 75% training data and 25% test data. The hyperparameters used were ntree = 100, learning rate = 0.03, and max-depth = 9. CatBoost also demonstrated the best consistency in avoiding overfitting across all tested scenarios. LightGBM and XGBoost performed lower due to their sensitivity to small-dimensional data during training, leading to inconsistent performance when applied to different proportions of training and test data. |
URI: | http://repository.ipb.ac.id/handle/123456789/154592 |
Appears in Collections: | UT - Statistics and Data Sciences |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
cover_G1401201014_.pdf Restricted Access | Cover | 724.47 kB | Adobe PDF | View/Open |
fulltext_G1401201014.pdf Restricted Access | Fullteks | 7.47 MB | Adobe PDF | View/Open |
Lampiran_G1401201014.pdf Restricted Access | Lampiran | 569.21 kB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.