RANDOM FOREST DAN MULTIVARIATE ADAPTIVE REGRESSION SPLINES (MARS) UNTUK KLASIFIKASI KEMANDIRIAN PADA ANAK USIA DINI DI KECAMATAN BOJONEGORO

Abstract

ABSTRACT

Independence is a situation where someone tries to stand alone in the sense of not
relying on others in a decision and is able to carry out life's tasks with full responsibility. There
are so many factors that influence independence, including parenting, age of the child, order of
children, self-esteem of children, genes, status of mother workers, and so forth. In this study
wanted to find out the factors that influence the independence of early childhood with the MARS
method and improve the accuracy of classification using the resampling method namely random
forest and a combination of MARS and random forest. The formation stage of the MARS model
begins by conducting trial and error on the maximum base function (BF), maximum interaction
(MI), and minimum number of observations between knots (MO) to obtain optimal values with
minimum GCV values. The next step is to take n samples with resampling techniques with
returns to obtain a new dataset. From the results of the analysis, the best MARS model was
obtained when the combination BF = 22, MI = 3 and MO = 1 with GCV values of 0.142. Of the
11 variables analyzed, 5 variables entered into the best MARS model, namely parenting, order
of children in the family, habits of children, social characteristics of children, and maternal
status. The accuracy of children's independence classification in Bojonegoro sub-district with
the MARS method is 77.39% with a random forest method of 95.95% and by the random forest
MARS method of 89.86%. So that the best classification accuracy is obtained using the Random
Forest method.
Keywords : MARS, random forest, independence of children