Please use this identifier to cite or link to this item: http://repository.ipb.ac.id/handle/123456789/110742
Title: Perbandingan Pengklasifikasian Metode Support Vector Machine dan Random Forest (Kasus Perusahaan Kebun Kelapa Sawit)
Authors: Rizki, Akbar
Soleh, Agus M
Achmad, Nabila Destyana
Issue Date: 2022
Publisher: IPB University
Abstract: Kelapa sawit adalah salah satu komoditi unggulan yang menopang perekonomian di Indonesia. Salah satu perusahaan yang bergerak di sektor perkebunan kelapa sawit memiliki 146 unit kebun kelapa sawit. Pengoptimalan produksi kelapa sawit sangat penting dilakukan sehingga diperlukan pengklasifikasian status unit kebun. Pengklasifikasian bertujuan untuk memprediksi unit kebun baru dan melihat peubah yang paling penting dalam proses pemodelan. Peubah yang digunakan adalah status unit kebun sebagai peubah respon dan sembilan peubah penjelas yaitu luas panen, curah hujan, buah normal, produksi tandan buah segar, brondolan, produksi, prestasi panen, luas kelompok pusingan panen, dan tenaga kerja. Proses pengklasifikasian dilakukan menggunakan metode Support Vector Machine (SVM) dan random forest untuk melihat metode mana yang paling baik. Data dibagi menjadi 80% data latih dan 20% data uji dengan sepuluh kali iterasi sehingga dihasilkan sepuluh model pada setiap metode. Evaluasi hasil model dilakukan dengan membandingkan nilai akurasi, skor F1, dan nilai Area Under Curve (AUC). Hasil pemodelan menunjukkan bahwa metode random forest memiliki performa yang lebih baik dibandingkan dengan metode SVM. Nilai rataan performa metode random forest pada akurasi, skor F1, dan AUC berturut-turut yaitu 90%, 86%, dan 89%. Peubah prestasi panen, brondolan, luas panen, curah hujan, dan luas kelompok pusingan panen adalah peubah penting yang berkontribusi lebih dari 10% dalam model. Hasil penelitian digunakan untuk proses evaluasi dan pengembangan perusahaan sawit dengan memerhatikan hasil peubah penting yang memengaruhi produktivitas dan hasil prediksi unit kebun baru.
Palm oil is one of the leading commodities that support the economy in Indonesia. One of the companies engaged in the oil palm plantation sector has 146 units of oil palm plantations. It is very important to optimize oil palm production, so it is necessary to classify the status of plantation units. Classification aims to predict new plantation units and find the most important variables in the modeling process. The variables used were the status of the garden as a response variable and nine explanatory variables, namely harvested area, rainfall, percentage of normal fruit, fresh fruit bunches production, oil palm loose fruits, production, harvest job performance, harvesting rotation, and farmers. The classification process is carried out using the Support Vector Machine and Random Forest methods to find which method is the best. The data is divided into 80% training data and 20% test data with ten iterations so that ten models are produced for each method. Comparing accuracy value, F1 score, and Area Under Curve (AUC) to evaluate the model. The modeling results show that the random forest method has better performance than the SVM method. The random forest has an average of accuracy, F1 score, and AUC, respectively, is 90%, 86%, and 89%. Variables of harvest job performance, oil palm loose fruits, harvested area, rainfall, and harvesting rotation are important variables that contribute more than 10% of the model. The results of the research are used for the evaluation and development process of oil palm companies by taking into account the result of important variables that affect productivity and predictive results of new plantation units.
URI: http://repository.ipb.ac.id/handle/123456789/110742
Appears in Collections:UT - Statistics and Data Sciences

Files in This Item:
File Description SizeFormat 
Cover, Lembar Pengesahan,Prakata, Daftar Isi.pdf
  Restricted Access
Cover967.23 kBAdobe PDFView/Open
G14170026_Nabila Destyana Achmad.pdf
  Restricted Access
Full text1.73 MBAdobe PDFView/Open
Lampiran.pdf
  Restricted Access
Lampiran806.35 kBAdobe PDFView/Open


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