A method for predicting postpartum depression via an ensemble neural network model.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computational Psychiatry

Background:

• Postpartum depression (PPD) significantly impacts families and society, necessitating early detection.
• Existing machine learning models for PPD prediction face challenges in achieving high accuracy and interpretability.

Purpose of the Study:

• To design an interpretable ensemble neural network model for accurate PPD prediction.
• To combine Fully Connected Neural Network (FCNN) and a Neural Network with Dropout (DNN) for enhanced performance.

Main Methods:

• Developed an ensemble model integrating FCNN for interpretability and DNN with Dropout for generalization.
• Determined model weights based on training accuracy and Dropout values.
• Ensured model stability by not solely relying on Dropout for overfitting prevention.

Main Results:

• Achieved high performance metrics: 0.933 accuracy, 0.958 precision, 0.939 recall, 0.948 F1-score.
• Outperformed 10 classic machine learning classifiers in accuracy, precision, recall, and F1-score.
• Demonstrated high stability across different dataset split ratios.

Conclusions:

• The proposed model significantly enhances PPD prediction accuracy and interpretability.
• Results offer guiding suggestions for clinicians and postpartum women.
• Future work includes expanding to predict other diseases and developing an online prediction platform.