DIGITAL BEHAVIORAL PATTERN MODELING FOR ANXIETY RISK PREDICTION

Solokhov Timur Damirovich

The increasing prevalence of anxiety disorders necessitates the development of new, scalable methods for pre-clinical screening. One of the most promising avenues is the analysis of digital behavioral patterns – persistent traces of human interaction with smartphones and online platforms that form a digital phenotype. This article aims to systematize and analyze modern methods of mathematical modeling of such patterns for building predictive models of individual risk for developing anxiety symptoms. The work reviews sources of relevant digital data, including passive smartphone sensing (mobility metrics, circadian rhythms) and social media activity. The primary focus is on the application of nonlinear dynamics tools, such as Recurrence Quantification Analysis and time-series entropy calculation, to identify chaos and regularity disturbances in behavior. Concurrently, the effectiveness of machine learning algorithms (gradient boosting, support vector machines) for binary classification tasks is investigated. The results demonstrate that complex models integrating heterogeneous digital features achieve high predictive accuracy (AUC-ROC up to 0.89). The article discusses in detail the key technological and ethical challenges related to ecological validity, model interpretability, and user privacy protection. It is concluded that mathematical modeling of digital traces holds significant potential for creating tools in preventive psychiatry, if ethical and clinically validated protocols for their application are developed.

Keywords:mathematical modeling, digital biomarkers, behavioral patterns, anxiety disorders, predictive models, machine learning, passive sensing, digital phenotype.