Development of Spatial Model and Flexible Model for Prediction of Low Birth Weight Events in East Java Province
Waego Hadi Nugroho (a), Agus Dwi Sulistyono (b*), Atiek Iriany (a), Novi Nur Aini (c)

(a) Department of Statistics, Faculty of Mathematics and Natural Sciences, Universitas Brawijaya
(b) Department of Socio-Economy Fisheries and Marine, Faculty of Fisheries and Marine Sciences, Universitas Brawijaya *agusdwistat[at]ub.ac.id
(c) Student at Magister Program of Statistics, Faculty of Mathematics and Natural Sciences, Universitas Brawijaya

Abstract

The incidence of Low Birth Weight (LBW) in Indonesia is still relatively high. Based on UNICEF data, Indonesia ranks 78th in the world in LBW cases, with a percentage of 10% of birth rates in Indonesia. Based on data from the Badan Pusat Statistik (BPS), the incidence of LBW in East Java Province is still relatively high at 20,836 people in 2016 and as many as 14,882 people in 2017. Many factors cause LBW, especially during pregnancy. Research on LBW about the risk factors has been carried out by several researchers, but the exposure of the study is still limited with tables, diagrams, or graphs as well as linear regression models or other univariate models. The purpose of this study is to establish a location-based LBW prediction model. This research was conducted using a Geographically Weighted Regression (GWR) model based on location in the City / District in East Java and a flexible model with a deep learning approach. Endogenous variables (Y) used are LBW events in East Java and exogenous variables are Percentage of Early Marriage (X1), Human Development Index (X2), Number of Health Facilities (X3), K1 Visit (X4), K4 Visit (X5), Consumption of Fe 30 (X6), and Consumption of Fe 90 (X7). Based on the results of the analysis using the GWR model, global equation models are obtained

and local models as many as 38 models with R2 = 82.06%. While based on the results of the analysis with a flexible model with a deep learning approach, the model is obtained

with R2 = 84.8%. From this study, it can be concluded that the GWR model and the flexible model have relatively the same level of accuracy. However, the flexible model can show the nonlinear effect of the variables X1 and X3.

Keywords: Low Birth Weight, Geographically Weighted Regression, machine learning, spatial model

Topic: HEALTH, NUTRITION AND MEDICAL MICROBIOLOGY