Klasifikasi Multikelas Support Vector Machine dengan Hibrida Directed Acyclic Graph One Vs One dan Rest Vs Rest pada Klasifikasi Tingkat Obesitas

Naufal, Daffa Ahmad
Nadeak, Christyan Tamaro
Rassiyanti, Linda
Farid, Fajri

MDP Student Conference
Vol 5 No 2 (2026): The 5th MDP Student Conference 2026

This research is focused on analyzing how well different multiclass Support Vector Machine (SVM) classification methods can predict obesity levels. It also presents a new hybrid Directed Acyclic Graph Rest-vs-Rest (DAG-RvR) method as a better option. The study utilizes a dataset called the Obesity Risk Prediction Cleaned, which has information on seven different obesity categories. The methods being assessed include One-vs-One (OvO), One-vs-Rest (OvR), DAG-One-vs-One (DAG-OvO), and the new DAG-RvR method. For fine-tuning the parameters, GridSearchCV and the RBF kernel were used. The findings reveal that DAG-RvR achieves an accuracy of 0.91, which is similar to OvO and DAG-OvO, but it trains much quicker, taking just 0.3422 seconds. Even though its precision, recall, and F1-score are a bit lower than the pairwise methods, DAG-RvR still maintains reliable multiclass performance. In summary, this method strikes a good balance between achieving high accuracy and being efficient in computations.