Graphical Causal Structure and Machine Learning Models in the Study of Infant Health and Mortality

The rise of medical and health data in digitized database storage allows computer scientists and researchers to design highly accurate predictive models using machine learning. However, the dependency of models on correlation fails to establish explainable causal effects, and models show lower performance on never-before-seen data despite their high predictive and inferential accuracy on the training–testing data and the likeness. There is a lack of causal explanation among the features and target of the data. These issues give rise to challenges of adaptability and cause-effect explainability. This paper proposes a model that analyses the effect of health policies, considering health policy and observations on infants as features, whereas infant death is the target. The structural causal model (SCM) establishes the cause-effect relationship between the features and the target. The causal features are highly emphasized and considered in feature selection with proper justification. The infant deaths are predicted using the machine learning models after the causal model establishes and validates the graphical causal structure. The model articulates the graphical relationship between the features and the target in terms of the causal graph (Directed Acyclic Graph) and fitness of the model, along with the prediction performances of the machine learning models.