Towards Machine Learning Approaches for Predicting Risk Level of Postpartum Depression
T. H. K. R. Prabhashwaree, N. Mihirini Wagarachchi
Abstract
This Postpartum depression (PPD) is approaching epidemic rates in many South Asian countries. It occurs in some mothers after giving childbirth because of changes in their physical, behavioral, and emotional development. The main objective of this research is to identify factors that reason for PPD based on the mother’s family, social background, and other data related to the status of the mother and develop a model to predict postpartum depression risk levels. Here, based on a postnatal period of Sri Lankan mothers at 6 months, risk levels have been classified into 4 classes mild, moderate, severe, and profound using the Edinburgh Postpartum Depression Scale (EPDS). After reviewing past literature has identified Feed-Forward Neural Network (FFANN), Adaptive Neuro-Fuzzy Inference System, Genetic Algorithm (ANFIS - GA), Random Forest (RF), and Support Vector Machine (SVM) best for building the proposed models. Finally, supposed to identify which model has good performance when predicting depending on the model’s performance. After model training and testing, as classification and regression types of models, the FFANN model (97.08% accuracy) and the ANFIS - GA model (testing error: 0.0496) have good performance. Finally, comparing the performance of both models for predicting PPD risk levels, it is concluded that FFANN has the best performance with multi classification. It has given great help to identify more influencing factors for PPD.