Postpartum Depression Risk Prediction using Hybrid Feature Selection and Ensemble Learning

Postpartum depression (PPD) is a common mental
health disorder that occurs in women following childbirth, which
often results in negative maternal and infant outcomes. Early
detection is still a challenge because of the complex interaction of
demographical, clinical and psychological factors. This research
purpose is to develop a powerful machine learning framework, to
accurately predict PPD risk. We used a combination of Chisquare,
mutual information and Random Forest importance to
select the most informative 25 features out of an initial set of 60
variables using a hybrid feature selection. Multiple machine
learning models were trained such as Logistic Regression,
Random Forest and Multi-Layer Perceptron (MLP) and their
ensemble was implemented for better predictive performance.
The proposed ensemble performed better with an accuracy of
91.3%, F1-score of 90.4%, and an AUC of 0.94, which is better
than individual models and numerous state-of-the-art methods.
These findings show that the hybrid feature selection combined
with ensemble learning can offer a reliable and interpretable tool
for the detection of early postpartum depression that can lead to
an earlier intervention process and contribute to better maternal
mental health outcomes.
Keywords: Postpartum Depression, Machine Learning, Feature
Selection, Ensemble Learning, Predictive Modeling, Edinburgh
Postnatal Depression Scale