MODELING AND EVALUATING THE LEVEL OF PSYCHO-EMOTIONAL LOAD OF USERS IN THE DIGITAL ENVIRONMENT

Abstract

The article addresses an urgent scientific and practical problem of increasing the objectivity and accuracy of assessing the psycho-emotional state of users in the digital environment by applying classification and clustering algorithms. The introduction substantiates the contradiction between the growing level of user digital distress and the limitations of traditional diagnostic methods, which determines the need to develop mathematical models based on digital traces. At the data preparation stage, a dataset of 5,000 observations with 25 parameters covering digital behavior, physiology, and mental health was cleaned, and a preprocessing pipeline was formed using One-Hot encoding of categorical variables and z-score standardization of numerical features. The authors proposed and mathematically formalized an original integrated Emotional Load Index (IEH), which balances the weight coefficients of risk factors and compensatory body resources proportionally to their correlation with stress levels. Using a multiple linear regression model (R² = 0.9389), it was proven that the primary trigger of distress is the duration of social media use rather than total screen time outside social platforms. Based on quantile division of stress levels into three classes (Low, Mid, High), a comparative analysis of Naive Bayes, Decision Tree, and SVM classification models with an RBF kernel was conducted. The results of 5-fold cross-validation demonstrated that the SVM RBF model possesses the highest generalization ability and robustness against overfitting (Accuracy = 0.9856). Applying k-means clustering (k=2, Silhouette score = 0.42) combined with visualization in the Principal Component Analysis (PCA) space allowed identifying two behavioral profiles: "Resourceful" and "Digital Risk". The practical value lies in creating a foundation for developing software modules for real-time automated monitoring of user mental health.