Selection of the optimal basis function on nutritional status of toddlers in Southeast Sulawesi using continue biresponses nonparametric regression model using MARS

The nonparametric regression method is one of the statistical analysis methods used when the shape of the data pattern and regression function is unknown. Multivariate adaptive regression splines (MARS) is one of the nonparametric regression methods that can overcome data problems with many predictor variables 3 - 20 and the number of samples 50 - 1000. If there are two response variables that are correlated, the model used is continue biresponses MARS. To accommodate the correlation between responses, the estimated regression function contains a weight matrix in the form of error covariance. The continue biresponses MARS model was used on the nutritional status of toddlers in North Sulawesi by involving two response variables: stunting and wasting. This study used sub-districts as the research unit, total 222 sub-districts. The continued biresponses MARS estimation results provided good accuracy with an R² value of 60.25% and a root mean square error (RMSE) of 7.35. Validation results with root mean square error cross validation (RMSEV) of 7.83 showed that the estimation results were valid enough to describe the phenomenon of stunting and wasting. The validation also shows that the prevalence of stunting and wasting that includes the correlation between responses results in a better prediction rate than those that do not involve correlation.