// <![CDATA[PERBANDINGAN METODE DECISION TREE, RANDOM FOREST, DAN GRAPH CONVOLUTIONAL NETWORK PADA KLASIFIKASI TINGKAT OBESITAS]]> 0404057502 - Youllia Indrawaty Nurhasah, ST., MT. Dosen Pembimbing 1 HENSHAMMI ADHA FERNANDI / 152021211 Penulis
Penelitian ini bertujuan untuk membandingkan performa metode Decision Tree, Random Forest, dan Graph Convolutional Network dalam klasifikasi tingkat obesitas menggunakan dataset gabungan dari UCI Machine Learning Repository dan data primer puskesmas di Kota Bandung. Metode penelitian melibatkan preprocessing data dengan label encoding dan normalisasi, diikuti pembagian data menjadi data latih dan uji, serta penerapan teknik semi-supervised learning untuk Pseudo-Labeling dan struktur graf untuk Graph Convolutional Network. Hasil pengujian menunjukkan bahwa Random Forest mencapai akurasi tertinggi sebesar 95,45%, diikuti oleh Kombinasi Random Forest dan Pseudo-Labeling menghasilkan akurasi 94,50%. Decision Tree dengan Pseudo-Labeling sebesar 94,02% dan Decision Tree saja sebesar 92,58%, sedangkan Graph Convolutional Network hanya mencapai 68,04% karena keterbatasan struktur graf dan tuning hyperparameter. Penelitian ini menegaskan keunggulan metode ensemble seperti Random Forest dalam menangani klasifikasi data obesitas. This study aims to compare the performance of Decision Tree, Random Forest, and Graph Convolutional Network methods in obesity classification using a combined dataset from the UCI Machine Learning Repository and primary data from community health centers in Bandung. The research method involves data preprocessing with label encoding and normalization, followed by dividing the data into training and testing data, and applying semi-supervised learning techniques for Pseudo-Labeling and graph structures for Graph Convolutional Network. The test results show that Random Forest achieved the highest accuracy of 95.45%, followed by the combination of Random Forest and Pseudo-Labeling, which yielded an accuracy of 94.50%. Decision Tree with Pseudo-Labeling achieved 94.02% and Decision Tree alone achieved 92.58%, while Graph Convolutional Network only reached 68.04% due to limitations in graph structure and hyperparameter tuning. This study highlights the superiority of ensemble methods like Random Forest in handling obesity data classification.