Performance Optimization of Support Vector Machine with SMOTE for Multiclass Stunting Prediction in Sumedang District, Indonesia

The percentage of stunting toddlers in Sumedang Regency is the highest compared to other nutritional problems. Stunting imposes a significant risk to the future quality of human resources. This study explores the performance of the Support Vector Machine (SVM) algorithm in predicting the stunting status of toddlers in Tanjungmedar Subdistrict, the region with the highest incidence of stunting cases in Sumedang Regency in 2020. The testing uses RapidMiner software and applies the Synthetic Minority Oversampling Technique (SMOTE) to overcome the imbalanced dataset so that the resulting performance can be optimized. Accuracy, precision, recall, and F1-score are measured in performance evaluation using a confusion matrix. The findings demonstrate that SMOTE might adjust the distribution of target classes in the dataset to maximize the SVM algorithm's performance. At the start of the test, the SVM model produced an accuracy of 85.10%. After applying SMOTE, the accuracy of the SVM model increased to 89.08%. The F1-score also increased for each class, except for the Normal class, which decreased slightly. These results demonstrate the suitability of SVM combined with SMOTE for health-related multiclass classification tasks, especially in imbalanced public health datasets, contributing to the advancement of applied machine learning in healthcare informatics.