Postpartum Depression Identification: A Novel Approach Employing ABC Algorithm and Off-Policy Proximal Policy Optimization
Abstract
Postpartum depression (PPD) is a significant public health issue, impacting approximately 12% of new mothers and adversely impacting the health of mothers and children. Despite its prevalence, many women with PPD lack treatment due to identification challenges. The motivation for this paper is to introduce a novel automated detection model for PPD to solve the problems of unbalanced classification and sensitivity to initial weights in previous approaches. Our model uses an off-policy proximal policy optimization (Off-Policy PPO) algorithm to solve the class imbalance problem by adjusting the reward mechanism during the training process. This is to enhance the accuracy of minority-class identification. This approach converts the classification procedure into a problem involving sequential decision-making inside the framework of an artificial neural network (ANN). Here, each correct classification action, especially of the minority class, receives increased rewards, enhancing the focus of the model on accurately detecting at-risk individuals. Furthermore, the artificial bee colony (ABC) algorithm is used for its proficiency in optimizing initial weight settings, which is crucial for navigating complex, high-dimensional spaces effectively. The performance of our model was rigorously tested employing 4313 samples and data from the population-based Biology, Affect, Stress, Imaging and Cognition during Pregnancy and the Puerperium (BASIC) research carried out in Uppsala, Sweden, between 2009 and 2018. The proposed model achieved an accuracy of 0.8928, which is superior to existing models. The proposed model could improve health outcomes for mothers and their children by enabling more targeted and effective treatment strategies.