Statistical Modeling of Under-Five Child Survival in the Lake Region Economic Bloc, Kenya
John Paul Wasiro *
Department of Statistics and Actuarial Sciences, Jomo Kenyatta University of Agriculture and Technology, Nairobi, Kenya.
Oscar Ngesa
Department of Mathematics Statistics and Physical Sciences, Taita Taveta University, Taita Taveta, Kenya.
Herbert Imboga
Department of Statistics and Actuarial Sciences, Jomo Kenyatta University of Agriculture and Technology, Nairobi, Kenya.
*Author to whom correspondence should be addressed.
Abstract
Under-five mortality remains a major public health challenge in Kenya despite substantial improvements in child survival over recent decades. The Lake Region Economic Bloc (LREB) continues to experience disproportionately high childhood mortality because of persistent socioeconomic inequalities, infectious diseases, and unequal access to healthcare. Although conventional survival models have been widely applied to identify determinants of child survival, evidence comparing their performance with that of machine learning survival models in this setting remains limited. This study modelled under-five survival in the LREB using secondary data from the 2022 Kenya Demographic and Health Survey (KDHS), comprising 3,894 children. Survival time was measured from birth to death before the fifth birthday or censoring at the survey date. Conventional survival models, including the Cox proportional hazards, Weibull accelerated failure time (AFT), and exponential AFT models, were compared with random survival forest (RSF), extreme gradient boosting survival (XGBoost Survival), and deep survival neural network (DeepSurv) models. Model performance was evaluated using the concordance index (C-index), integrated Brier score (IBS), time-dependent area under the receiver operating characteristic curve (time-dependent AUC), Akaike information criterion (AIC), Bayesian information criterion (BIC), and log-likelihood. Under-five survival probability decreased with age. Place of residence and household socioeconomic status were the most consistent determinants of survival: urban residence was associated with a lower mortality risk, whereas children from poorer households had poorer survival outcomes. The Weibull AFT model achieved the best predictive performance (C-index = 0.5357), followed by the Cox proportional hazards and exponential AFT models. DeepSurv, RSF, and XGBoost Survival showed comparatively lower predictive accuracy. The findings indicate that the conventional survival models adequately represented the survival process in the available data. The Weibull AFT model was therefore the most suitable evaluated framework for modelling under-five survival in the LREB. Strengthening rural health systems, reducing socioeconomic inequalities, expanding access to maternal and child healthcare, and incorporating richer longitudinal and contextual data may improve future survival modelling and support targeted child-health interventions.
Keywords: Under-five survival, survival analysis, Weibull accelerated failure time model, machine learning survival models, cox proportional hazards model, lake region economic bloc, Kenya Demographic and Health Survey