Predicting Women with Postpartum Depression Symptoms Using Machine Learning Techniques

Abstract

Being pregnant and giving birth are big life stages that occur for women. The physical and mental effects of pregnancy and childbirth, like those of many other fleeting life experiences, have the significant potential to influence a mother’s overall health and well-being. They have also been known to trigger Postpartum Depression (PPD) in many cases. PPD can be exhausting for the mother and it may have a negative impact on her capacity to care for herself and her kid if it is not treated. For this reason, in this study, initially, physiological questionnaire Edinburgh Postnatal Depression Scale (EPDS) data were collected from delivered mothers for one week, the score was evaluated by medical experts, and participants with PDD symptoms were identified. As a part of multistage progress, further, follow-up was carried out by collecting the Patient Health Questionnaire-9 (PHQ-9), Postpartum Depression Screening Scale (PDSS) questionnaires for the above-predicted participants until six weeks. As the second step, correlated risk factors with PPD symptoms were identified using statistical analysis. Finally, data were analyzed and used to train and test machine learning algorithms in order to predict postpartum depression from one to six weeks. The extremely Randomized Trees (XRT) algorithm with (Background Information + PHQ-9 + PDSS) data offers the most accurate and efficient prediction. Pregnant women with these features could be identified and treated properly. Moreover, it reduces prolonged complications and remains cost-effective in future clinical models.

Keywords

• postpartum depression (PPD)
• psychometric questionnaire (EPDS, PDSS, PHQ-9)
• depression analysis
• class imbalance problem
• classification algorithms