JMIR Publications

ABSTRACT

Background:

Machine Learning is one of the important methods for disease prediction

Objective:

This study combined different machine learning methods to establish four postnatal depression prediction models, and compared their performance to determine the prediction model with best performance .

Methods:

A total of 1126 pregnant women were enrolled in two women and children’s hospitals. A questionnaire survey was conducted on pregnant women in early pregnancy, second trimester, third trimester. Under the different Feature Selected methods of Expert knowledge method and the Filter Feature Selection algorithm based on random forest, four postnatal depression prediction models were generated by Support Vector Machine and Random Forest algorithm respectively. After optimizing the parameters of the model, a series of indicators are used to evaluate the predictive performance of each model, in order to find the best model.

Results:

We found that our four postnatal depression prediction models, models built using Random Forest modeling and the filter feature selection algorithm based on random forest feature selection has the best performance. The sensitivity of the model constructed by Support Vector Machine is slightly higher than that of the random forest approach. The importance of psychological resilience is found to be higher than other model factors.

Conclusions:

The postnatal depression prediction model established by Support Vector Machine proves the value when the sample content is smaller. The two Feature Selection methods have little effect on the performance of postnatal prediction model. Psychological resilience may be an important factor affecting the occurrence of postpartum depression. This application of machine learning may provide a sensitive approach for the the prediction and prevention of postpartum depression.