Childhood circumstances may impact senior health, prompting this study to introduce novel machine learning methods to assess their individual and collective contributions to health inequality in old age. Using the US Health and Retirement Study (HRS) and the China Health and Retirement Longitudinal Study (CHARLS), we analyzed health outcomes of American and Chinese participants aged 60 and above. Conditional inference trees and forest were employed to estimate the influence of childhood circumstances on self-rated health (SRH), comparing with the conventional parametric Roemer method. The conventional parametric Roemer method estimated higher IOP in health (China: 0.039, 22.67% of the total Gini coefficient 0.172; US: 0.067, 35.08% of the total Gini coefficient 0.191) than conditional inference tree (China: 0.022, 12.79% of 0.172; US: 0.044, 23.04% of 0.191) and forest (China: 0.035, 20.35% of 0.172; US: 0.054, 28.27% of 0.191). Key determinants of health in old age were identified, including childhood health, family financial status, and regional differences. The conditional inference forest consistently outperformed other methods in predictive accuracy as measured by out-of-sample mean squared error (MSE). The findings demonstrate the importance of early-life circumstances in shaping later health outcomes and stress the early-life interventions for health equity in aging societies. Our methods highlight the utility of machine learning in public health to identify determinants of health inequality.