Comparison of XGBoost and LightGBM Methods with Random Oversampling for Obesity Level Classification

Rafika, Chesa Saskia (2026) Comparison of XGBoost and LightGBM Methods with Random Oversampling for Obesity Level Classification. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text 22081010211.-cover skripsi.pdf Download (1MB)
	Text (Bab I) 22081010211.-bab1.pdf Download (22kB)
	Text (Bab II) 22081010211.-bab2.pdf Restricted to Repository staff only until 14 July 2028. Download (585kB)
	Text (Bab III) 22081010211.-bab3.pdf Restricted to Repository staff only until 14 July 2028. Download (1MB)
	Text (Bab IV) 22081010211.-bab4 .pdf Download (7MB)
	Text (Bab V) 22081010211.-bab5.pdf Download (126kB)
	Text (Daftar Pustaka) 22081010211.-daftarpustaka.pdf Download (131kB)
	Text 22081010211.-lampiran.pdf Restricted to Repository staff only until 14 July 2028. Download (285kB)

Abstract

Obesity is a health issue related to several aspects of individual condition and daily behavior, including physical characteristics, food consumption habits, physical activity, and lifestyle patterns. This study was conducted to evaluate the performance of XGBoost and LightGBM in classifying obesity levels and to analyze the use of Random Oversampling (ROS) for handling unequal class distribution. The research used the Dhaka Obesity Dataset, which initially contained 2,182 respondents with 17 attributes. The data preparation process involved selecting adult respondents by removing data under 18 years old, reconstructing the target labels based on the Asia Pacific Body Mass Index (BMI) classification adopted by the Indonesian Ministry of Health, cleaning the dataset, transforming categorical variables into numerical form, and dividing the data using stratified train test split. After preprocessing and duplicate removal, the final dataset consisted of 2,125 records. The experimental stage was carried out by testing several train test split ratios, applying stepwise manual hyperparameter tuning, and comparing model performance under baseline and ROS scenarios. Model evaluation was performed using accuracy, macro precision, macro recall, macro F1 score, balanced accuracy, confusion matrix, classification report, and train test gap. The results showed that ROS improved the performance of both XGBoost and LightGBM, particularly by producing more balanced classification results across obesity classes. Among all evaluated models, LightGBM with ROS achieved the highest performance, with an accuracy of 0.967059, macro precision of 0.955319, macro recall of 0.942773, macro F1 score of 0.948016, and balanced accuracy of 0.942773. Therefore, LightGBM with ROS was selected as the most suitable model for obesity level classification in this study.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Puspaningrum, Eva Yulia	NIDN0005078908	evapuspaningrum.if@upnjatim.ac.id
Thesis advisor	Mumpuni, Retno	NIDN0016078703	retnomumpuni.if@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA75 Electronic computers. Computer science

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

chesa saskia rafika

Date Deposited:

14 Jul 2026 08:22

Last Modified:

14 Jul 2026 08:31

URI:

https://repository.upnjatim.ac.id/id/eprint/55379