Model Optimasi SVM-GSBE dalam Menangani High Dimensional Data Stunting Kota Samarinda

Abstract

Stunting has become a widely discussed health issue in Indonesia, par-ticularly in Samarinda City, which recorded a prevalence of 12.7% in 2023, making it the highest in East Kalimantan Province. The use of data mining techniques becomes crucial in overcoming the challenges of high dimensional data, such as computational complexity, the risk of overfitting, and visualization difficulties. This study aims to enhance the accuracy of Support Vector Machine optimization models using Grid Search and Backward Elimination feature selection (SVM-GSBE) to handle high-dimensional data related to stunting in Samarinda City. The dataset used is sourced from Samarinda City Health Office in 2023, covering 26 community health centers with 21 attributes and a total of 150,466 records. The research methodology includes data collection, pre-processing, data partitioning using K-Fold Cross Validation, feature selection using Backward Elimination, and SVM model optimization with Grid Search. Features such as BB/U, ZS TB/U, ZS BB/U, ZS BB/TB, Height, and LiLA have proven to increase accuracy in stunting data classification. Evaluation results show that Grid Search successfully increased accuracy for Linear from 99.59% to 99.78%, Polynomial from 90.92% to 99.40%, RBF from 89.80% to 98.36%, and Sigmoid from 75.29% to 86.84%. This indicates that the SVM-GSBE model can effectively be used as a tool for early detection of stunting and to support health policies in Samarinda City.