Small Area Estimation in Health Sector by Support Vector Machine (SVM)

Abstract

TheGovernment of India has initiated many public health policies in rural and urbansetup.An impact assessmentof the programme is a prime factor to explore status of the programme. Different analytical stepswere used forassessment of various policiesat national level.Small area estimation by support vector machine isone of the advanced statisticalmethods to magnify theprogramme at national level presenting in 3D dimensionality. The method of survey conducted in small area has differing based on the objective of our interest. If the objective willfocus on narrowpoint, certain limited number of health indicators  willbeselected for surveypurpose in smaller area. But, our objective of interest is ofbroader perspective lack of limitation as we will be selecting larger area for survey. Usually, the health and social survey mainly dependson rationality, economy and time duration. Byminimizing the economy andtime duration;hence we will achieve goodsuccess in researchprogramme. Without this thumb rule of research hypothesis,  ourtested hypothesismay not yield good resultsfrom the experimentalarea. Keeping this in view, initiallysmall area will be selected with limited sample size, later on the research is magnified at populationlevel in different dimensions. A very limited research hypothesis was testedbased on small area estimationapproach mathematically in health sector. In this ample ofresearch gap.The present study we attempted to fill the research gap to explore new statistical methods applicable to small area estimation and we  have  demonstrated  robustness of the model by using  sensitivity  analysis vetted in real  health data sets.The study was conducted in selected states and Union territoriesof India. The selected area constitutes of418 districts which are administered bytheir respective State /Union territories. As per the official record, district level health planning involves 660rates with various relative comparative attributes. In each selected sites, various health domains secondary data sets were collected based on SOP.The collected data was modelled by using Mat labsoftware, the support vector machine learning (SVM) was used to estimate various health indicators and magnify the results at population level, In this study intervention,we have adopted  two-  stage sampling procedure for selection of 20 sub-centres in selected districts of south India, we visited all households and collected relevant information (datasets) regarding  recent births and socioeconomic data from selected respondents of every married women aged between 15–49 years. SVM was adopted to optimize thecensoredsurvey data on every married woman to know the general marital fertility (GMFR) rate. As per the analysis, the median general marital fertility (GMF) rate was285.22(63.25%) [Likelihood10.52, R2(%) =0.86, Akiake information 125.22].Our model results revealed that,the medianGeneral Marital Fertility Rate (GMFR) is simple estimation of number of rates of births (simulated selected socio-economic factors) in small area. Our demonstratedmodel willhelpus topredict various health outcomes and also supportthepolicy makers for implementation of public health policy at nationallevel to demystify the art of valid decisions. The SVMsummary willmeasure various generalized attributes and also reports small area variation inclusion with fertility rates and necessary intervention atsample level.