Optimization of Support Vector Machine Performance for Prediction of Stunting Status in Toddlers
Irfan Fadil (a*), Ucu Koswara (b)

a) Faculty of Information Technology, Sebelas April University
Jalan Angkrek Situ No. 19 Sumedang,Indonesia
*fadilirfan[at]unsap.ac.id
b) Faculty of Teacher Training and Education, Sebelas April University
Jalan Angkrek Situ No. 19 Sumedang, Indonesia

Abstract

Indonesia is one of the countries with various nutritional issues. Indonesia is still working hard to overcome nutrition issues, one of them is stunting. In Sumedang Regency, the percentage of stunted toddlers is the highest compared to other nutrition issues. Stunting is a major threat to the quality of human resources in the future. The purpose of this research is to optimize the performance of the Support Vector Machine (SVM) algorithm to produce a good model for classifying or predicting stunting status in toddlers. In this study, the performance of the SVM algorithm will be tested in predicting the stunting status of toddlers using data from Tanjungmedar District, which is the area with the most stunting cases in Sumedang Regency in 2020. Our tests use RapidMiner software and apply the SMOTE (Synthetic Minority Oversampling Technique) to overcome imbalanced datasets so as to optimize the resulting performance. Performance assessment uses confusion matrix to measure Accuracy, Precision, Recall, and F1-score. The results show that SMOTE can optimize the performance of the SVM algorithm by balancing the distribution of target classes on the dataset. At the beginning of testing, the SVM model can produce an accuracy of 85.10%. After applying SMOTE, the accuracy of the SVM model increased to 89.08%. Based on the results of the research conducted, it can be concluded that the SVM classifier with SMOTE optimization is appropriate to be used as a model for classifying or predicting stunting status in toddlers.

Keywords: Data Mining-Support Vector Machine-SMOTE-Stunting

Topic: Computational Science