Bi-Response Nonparametric Modeling Using Hybrid Multivariate Adaptive Regression Spline and Support Vector Regression (MARS-SVR)

Abstract

The HDI and poverty rate are some of the standards for the success of a country's development through efforts to improve the quality of human life. An assessment of the achievement of development success indicates that there are many important factors, so it is necessary to identify them. This study aims to examine the MARS and SVR modeling (single-stage) and also the hybrid MARS-SVR bi-response. One of the problems is that SVR cannot identify the relative importance of the explanatory variable when many potential variables are considered. Thus, the MARS method was adapted to identify the important variables used as explanatory variables in the SVR method through hybrid modeling. Bi-response modeling is applied on society welfare indicators as seen from the poverty rate and HDI in 34 provinces of Indonesia 2020. The data used are secondary data from BPS and the Indonesian Ministry of National Development Planning/BAPPENAS 2021. The estimation results of the weighted hybrid MARS-SVR bi-response provide good accuracy with RMSECV value of 0.0043. The model validation also shows good prediction results with RMSEPCV value of 0.618 and the correlation between actual testing and predicted as 98.365% for the first response and 99.946% for HDI (second response).