Multivariate Adaptive Fuzzy Clustering Means Regression Splines Model Using Generalized Cross-Validation (GCV) on Stunting Cases in Southeast Sulawesi

Regression modeling cannot be solved with a parametric approach if there is no information about the shape of the function, and there is no clear pattern of relationship between the response variable and the predictor variables. One approach that can be done is to use a nonparametric regression approach. Multivariate adaptive regression splines (MARS) are one of the nonparametric regression models that can accommodate additive effects and interaction effects between predictor variables in data modeling. MARS is a model capable of handling high-dimensional data, for sample sizes of 50–1000 and 3–20 predictor variables. This study uses the MARS model with fuzzy clustering means (FCM) which is then called the multivariate adaptive fuzzy clustering means regression splines model (MAFCMRS), creating a continuous model at knots by paying attention to aspects of heterogeneity. The results of the cluster validity test using the validity of the Xie and Beni index produced 5 clusters which will be formed into the multivariate adaptive fuzzy clustering means regression splines model. This model was used to model the prevalence of stunting in Southeast Sulawesi province using 222 district observation units. The results obtained from the 4 predictor variables used are number of pregnant women with chronic energy deficiency (CED), number of low birth weight of the baby, number of neonate visits, and number of health centers, all of which affected the prevalence of stunting at the 5% significance level.