Using explainable artificial intelligence for mapping health vulnerability: Interaction-based analysis of multiple...

Even as there are extensive genetic linkages across Latin America, local health risk is influenced by a host of interdependent factors that include (a) ethnic heterogeneity, (b) geographical isolation, and (c) disproportionate access to healthcare. The article presents a new explainable artificial intelligence (XAI) model for mapping and interpreting health vulnerability, combining several open-access datasets, such as disease prevalence, medical supply, and genetic ancestry profiles. We introduce a compound Interaction Index, as the product of ethnic diversity (E), inverted medical access (1 - M), and altitude (A), to quantify compounded structural and biological risk factors. Applying supervised learning models (F1 = 0.596 for SVM, F1 = 0.571 for gradient boosting, and logistic regression), in combination with unsupervised clustering and interpretable classification trees, we detect the high-risk regions with high diversity, low access, and mid-to-high altitude. This transparent and scalable methodology for equitable public health planning illuminates such 'clusters of vulnerability' which might remain hidden amid aggregate data.