Belief to Behavior Causal Graph Transformer (BB CGT): A Novel Machine Learning Framework for Predicting Breastfeeding Outcomes

Breastfeeding is a fundamental health component of the maternal and child health continuum, yet psychosocial perceptions have a major impact on its length and exclusivity. Emerging machine learning (ML) models are facilitating new ways to anticipate outcomes and develop culturally relevant interventions in line with Saudi Vision 2030. A cross-sectional study involving 394 mothers in Tabuk City, Saudi Arabia. Data were obtained utilizing structured Likert scale questionnaires that measured maternal constructs such as religious duty, partner support, concerns of body image, public acceptance and health literacy. Belief to Behavior Causal Graph Transformer (BB CGT) was deployed to encode the beliefs, build causal pathways, and foretell breastfeeding effects. Model performance was compared with logistic regression, random forest, and gradient boosting. Counterfactual simulations were also conducted to estimate the impacts of targeted interventions. Descriptive analysis showed strong endorsement of breastfeeding superiority, religious duty, and partner support, whilst body image concerns provided inconsistent responses. Breastfeeding exclusively at six months was positively related to religious duty, partner support, and public acceptance. The BB CGT model achieved a substantially higher accuracy (86 percent accuracy, AUROC 0.90, and F1 score 0.85) than other classical models. An analysis of attention found the strongest predictors to be partner support, religious duty, and immunity beliefs. Counterfactual simulations also indicated that increasing partner support could increase breastfeeding duration by 2.5 months, heighten health literacy and lowering body image concerns by 1.9 and 1.4 months respectively, with cumulative benefits of 3.8 months derived from combined therapy. Psychosocial beliefs play an important role in breastfeeding practices. The BB CGT model was designed to predict with accuracy and include actionable simulations relevant for culturally salient interventions to maximize maternal and child health within the framework of the Saudi Vision 2030.