Unveiling the prevalence and risk factors of early stage postpartum depression: a hybrid deep learning approach
Umesh Kumar Lilhore,Surjeet Dalal,Neetu Faujdar,Sarita Simaiya,Mamta Dahiya,Shilpi Tomar,Arshad Hashmi
DOI: https://doi.org/10.1007/s11042-024-18182-3
IF: 2.577
2024-01-26
Multimedia Tools and Applications
Abstract:A major psychological problem that numerous new mothers experience is postpartum depression (PPD). A woman's capacity to care for herself and her child may be hampered by the sadness, anxiety, and weariness it can bring on. Using information from their past medical treatment, lifestyles, and habits, deep learning can assist in identifying women who are at risk for PPD. By analyzing large amounts of data from various sources, deep learning models can help healthcare providers provide targeted support and interventions to women most likely to develop PPD. This can improve outcomes for both mothers and babies and help to reduce the overall burden of PPD on families and society. To identify and validate risk variables for PPD disease in women at an early stage, this investigation provides a CNN-BLSTM with a TL-based model. The hybrid model combines transfer learning (TL) with bi-directional long short-term memory (BLSTM) models and convolutional neural networks (CNNs) models. This model takes input data such as medical history, demographic information, and behavioural patterns and outputs the probability of the patient having PPD. The CNN component of the model mainly extracts the relevant features from the input data, while the LSTM component helps to analyze the sequential patterns in the data. By combining the strengths of both CNN and BLSTM, the model can effectively capture complex relationships within the input data and make accurate predictions. In the proposed model, we used it for transfer learning after Fine-tuning a pre-trained CNN + BLSTM model on a large dataset of non-PPD data to extract features. TL method helps to enhance the accuracy. We utilize an online PPD dataset which contains 1503 records of various women. The proposed model and well-known ML methods, i.e., Decision Tree, Random Forest, AdaBoost, LightGBM, XGBoost and CatBoost, were compared using various performance measuring parameters, i.e., Precision, Recall, F1-score, accuracy, and ROC_AUC. The proposed hybrid model achieved 99.6% accuracy and 99.7% precision which is higher than other methods.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering