Mental Resilience Among Women Using Sparse Autoencoder and Self-Training Ensemble

Mental health issues among women present significant challenges, often constrained by limited awareness and pervasive societal stigmas. This paper introduces an innovative approach employing Semi-Supervised Learning (SSL) and Ensemble Techniques to detect mental strength patterns within a dataset focused on participants' psychological functioning. We used CTGAN, an AI-driven method, to generate synthetic data to address data scarcity. This was followed by a combination of techniques, including Random Oversampling to augment the dataset and the application of a Synthetic Sparse Autoencoder (SSAE) to perform feature extraction. Principal Component Analysis (PCA) facilitated efficient dimensionality reduction, expediting model training. Building on this, we implemented a Self-Training Ensemble Model (STEM), integrating logistic regression, random forest classification, and naive Bayes classifiers. This ensemble approach, bolstered by labeled and sparse regularization, enhanced the model's ability to identify mental health indicators within the women-focused dataset. Our methodology demonstrated compelling results, achieving an accuracy of 0.779. This approach holds promise in augmenting mental health analysis, especially within datasets concerning women's psychological functioning, where data limitations pose significant challenges. The proposed method offers potential for early detection and targeted interventions, significantly impacting mental health care.