Objective:

To create and test a machine learning model for predicting postpartum depression (PPD) using comprehensive infant and maternal health indicators.

Approach:

• Study Design: A prospective study enrolling 273 postpartum women, collecting data on 44 demographic, obstetric, and clinical variables.
• Data Analysis: Participants were divided into training and validation sets after propensity-score matching. Feature selection was performed using LASSO regression, and nine machine learning algorithms were compared.
• Model Evaluation: Performance was assessed using Area Under Curve (AUC), calibration, and decision-curve analyses, with SHAP used for model interpretability.

Key Findings:

• LASSO regression identified four key predictors of PPD: unplanned mode of delivery, premature rupture of membranes, NRS pain score at 10 cm cervical dilation, and socioeconomic subclass.
• The random forest model achieved an AUC of 0.952 in the training set and 0.745 in the validation set.
• SHAP analysis indicated that unplanned delivery and high intrapartum pain were strong positive contributors to PPD risk, while higher socioeconomic status was protective.

Interpretation:

The interpretable random-forest model accurately predicted PPD six weeks postpartum, integrating explainable artificial intelligence with obstetric data.

Limitations:

• The study's generalizability may be limited due to its single-center design.
• Future studies should incorporate biological and psychosocial markers for improved applicability.

Conclusion:

The model provides a practical tool for individualized screening and early intervention for PPD.