MONITORING PTSD FACTORS IN WOMEN OF REPRODUCTIVE AGE USING SUPERVISED AND UNSUPERVISED MACHINE LEARNING METHODS

Abstract

Post-traumatic stress disorder (PTSD) has emerged as a critical public health conse- quence of the ongoing armed conflict between Ukraine and the russian federation, dispro- portionately affecting the mental health of civilian populations — particularly every woman of reproductive age. This study presents a rigorous statistical analysis of an anonymous population-based survey, employing a comprehensive suite of machine learning (ML) methodologies to develop a rapid screening framework for individuals at elevated PTSD risk. The analytical pipeline integrated both supervised and unsupervised learning approaches, encompassing classical logistic regression, Bayesian logistic regression, Random Forest, Stochastic Gradient Boosting, and Neural Network Classification. This multi-method framework enabled systematic deconstruction of the underlying data architecture and identification of key biopsychosocial predictors of PTSD onset. Unsupervised Random Forest clustering was subsequently applied to achieve objective sample segmentation based on identified predictive factors. The analysis yielded two statis- tically robust clusters demonstrating significant divergence in diagnosed PTSD prevalence (p < 0.001). The resulting predictive model confirms that PTSD risk stratification is feasible using demographic and behavioral markers augmented by a minimal set of targeted psycho- logical indicators — without recourse to clinical laboratory assessment. This operational simplicity renders the proposed framework directly deployable by social workers, primary care providers, and civil protection personnel operating under wartime resource constraints.