Investigation of infant mortality prediction methods and progression of its determinants in Uganda (1995-2016)

Abstract

This study looked at forecasting and prediction techniques for infant mortality as well as progression in the determinants of infant mortality from 1995 to 2016. The study specifically looked at how individual, household, and community factors contributed to infant mortality on a periodic basis between 1995 and 2016; assessed how these factors contributed to the differences in infant mortality between urban and rural areas over time; and evaluated various prediction and forecasting techniques using UDHS data sets. Using STATA 15 and EViews 5.0, the study was done at the univariate, bivariate, and multivariate (O.B. multi-level logistic decomposition; VECM; BVAR; and VAR). I used a multilevel decomposition of the mixed logistic regression model to measure the impact of grouped factors on time variation in the risks of infant death. The most effective algorithm for predicting infant mortality was found using Python. In the medium and long term, the effect differences accounted for a greater portion of the overall reduction in infant mortality. Factors at the household level accounted for the largest portion of the overall reduction in infant mortality. A higher percentage of the total reduction in infant mortality between 2001 and 2016 (82.4%) and 2001 and 2011 (31.0%) was generally attributable to the impacts of maternal characteristics (C), which include marital status, education level, birth interval, and intended pregnancy. To the contrary, proximate factors overall accounted for a larger share of the total difference in infant mortality due to compositional differences in the long run than the short run. Child, father, maternal, and household factors made up a larger share of the explained variance in the risk of infant mortality across communities (PCV) in rural areas. In urban settings, most of the variations were explained by community and proximate-level variables. The effect (C) and compositional (E) differences led to significant changes in infant mortality in both rural and urban areas in the short and long run. Irrespective of residence, the effect (C) differences represented a larger share of the total difference in infant mortality during the two periods. The CatBoost prediction algorithm performed best. Increased GDPP reduces neonatal mortality faster, and GDP reduces infant mortality faster. Short-run forecasting can be done using VAR and BVAR; long-run forecasting should employ VCEM. The government should fast-track health insurance bills, develop infant at-risk prediction software for community health workers, increase household disposable income, invest in health infrastructure, conduct preconception counseling, open child care centers for the first time, and support young mothers. It is therefore important policy to operate at both the community and household levels to correct such obstacles of infant mortality. Forecasting IMR and NMR should be based on VAR, and BVAR in SR and LR forecasts should be based on VECM.