Implementation of the Support Vector Machine Method for Early Detection of Stunting Based on Anthropometric Features

Abstract

Stunting describes chronic undernutrition during the growth and development period from the beginning of life. This situation is represented by a z-score for height for age (TB/A) less than -2 standard deviation (SD). The current method used to detect stunting in toddlers is to use KMS. The way to do this is to weigh the under-fives every month, the weighing results are recorded in the KMS, and between the points of weight from last month's weighing results and the results of this month's weighing are connected with a line. The series of child growth lines forms a child growth chart. This procedure is of course less effective. Based on these problems, the research conducted is the Implementation of the Support Vector Machine Method for Early Detection of Stunting Based on Anthropometric Features. The SVM method consists of a training process as system learning and testing to obtain classification results. The parameter tests carried out are lambda, complexity, and maximum iteration tests. The data used in this study were 90 data which were divided into 2 classes, namely stunting toddlers and normal toddlers. The SVM algorithm is a linear classification method, so it uses the kernel to deal with nonlinear data. The final results of this research produce the highest average accuracy of 86% λ = 10, C = 1, itermax = 200 and also use polynomial kernels. Comparison of the results of the classification of child stunting with the help of midwives shows that the system produces good accuracy