For the full paper, click this Link: https://arxiv.org/abs/2608.22809
Authors: Syed Shariar Alam Shuvo et al
Abstract
Postpartum depression (PPD) is a major maternal health challenge, particularly in low- and middle-income countries. This study proposes SAGE, a Stability-Aware Graph-Based Ensemble feature selection framework combined with local explainable AI and a genetically optimized artificial neural network (GA-ANN). Using data from 766 postpartum women, SAGE identified 16 robust predictors and achieved 87.96% accuracy, 86.32% F1 score, and 0.88 AUC. Key predictors included EPDS and PHQ-9 scores, feelings about motherhood, and abuse history. LIME-based explanations provided personalized, patient-level insights. Overall, SAGE offers an interpretable and scalable approach for early PPD risk prediction, particularly in resource-limited healthcare settings.
Introduction
Postpartum depression (PPD) is a serious maternal health condition, affecting about 17% of mothers globally and more than 19% in low- and middle-income countries. In Bangladesh, prevalence is particularly high due to economic hardship, limited mental health services, obstetric complications, and psychological and socioeconomic risk factors.
Machine learning offers strong potential for early PPD detection, but existing approaches often suffer from class imbalance, unstable feature selection, weak validation, and limited interpretability. These issues reduce clinical reliability and real-world usability.
To address these challenges, this study proposes SAGE, a Stability-Aware Graph-Based Ensemble feature selection framework combined with a genetically optimized artificial neural network (GA-ANN), GAN-based oversampling, and LIME explainability. Using data from 766 postpartum women in Bangladesh, the framework identifies stable and clinically meaningful predictors while providing patient-level risk explanations.
The study highlights psychological and socioeconomic factors as major predictors of PPD and presents an interpretable, robust approach for early risk identification in resource-limited healthcare settings.
For the full paper, click this Link: https://arxiv.org/abs/2608.22809
0 Comments