Detecting Post-Traumatic Stress Risk via ML Analysis of Dissociative Tendencies and Arousal Dysregulation

Objective: To evaluate the predictive utility of machine learning models in detecting post-traumatic stress risk by analyzing the complex interplay between dissociative tendencies and physiological arousal markers.Methods and Materials: A cross-sectional observational study was conducted with a sample of N=۱,۲۴۵ adult participants from the USA with documented trauma exposure. Data collection utilized the Posttraumatic Stress Disorder Checklist (PCL), the Dissociative Experiences Scale (DES), subjective hyperarousal scales, and wearable biometric sensors capturing electrodermal activity and heart rate variability during a standardized stress-reactivity paradigm. Extreme Gradient Boosting, Random Forest, and Support Vector Machine algorithms were trained and evaluated using stratified ten-fold cross-validation. Findings: The Extreme Gradient Boosting model demonstrated superior predictive performance, achieving an AUC=۰.۹۱, which significantly outperformed the Random Forest (AUC=۰.۸۸) and Support Vector Machine (AUC=۰.۸۲) models. Feature importance analysis revealed that derealization (۲۲.۴%), depersonalization (۱۸.۱%), and low heart rate variability during recovery (۱۵.۳%) were the most critical predictors of post-traumatic stress risk. Furthermore, a significant non-linear interaction demonstrated that objective physiological arousal strongly predicted risk primarily in highly dissociative individuals who underreported subjective distress. Conclusion: Machine learning models integrating objective physiological data with subjective dissociative measures offer a powerful, highly sensitive approach for detecting hidden post-traumatic stress risk. Objective: To evaluate the predictive utility of machine learning models in detecting post-traumatic stress risk by analyzing the complex interplay between dissociative tendencies and physiological arousal markers. Methods and Materials: A cross-sectional observational study was conducted with a sample of N=۱,۲۴۵ adult participants from the USA with documented trauma exposure. Data collection utilized the Posttraumatic Stress Disorder Checklist (PCL), the Dissociative Experiences Scale (DES), subjective hyperarousal scales, and wearable biometric sensors capturing electrodermal activity and heart rate variability during a standardized stress-reactivity paradigm. Extreme Gradient Boosting, Random Forest, and Support Vector Machine algorithms were trained and evaluated using stratified ten-fold cross-validation. Findings: The Extreme Gradient Boosting model demonstrated superior predictive performance, achieving an AUC=۰.۹۱, which significantly outperformed the Random Forest (AUC=۰.۸۸) and Support Vector Machine (AUC=۰.۸۲) models. Feature importance analysis revealed that derealization (۲۲.۴%), depersonalization (۱۸.۱%), and low heart rate variability during recovery (۱۵.۳%) were the most critical predictors of post-traumatic stress risk. Furthermore, a significant non-linear interaction demonstrated that objective physiological arousal strongly predicted risk primarily in highly dissociative individuals who underreported subjective distress. Conclusion: Machine learning models integrating objective physiological data with subjective dissociative measures offer a powerful, highly sensitive approach for detecting hidden post-traumatic stress risk.

کلیدواژه ها:

Post-Traumatic ، Stress Disorder ، Machine Learning ، Dissociation ، Arousal Dysregulation