The article explores the application of a systems approach and machine learning methods to forecast psychoemotional states based on digital activity in social networks. The study addresses the urgent need to assess the psychological impact of increasing user engagement with digital platforms by using quantitative and algorithmic tools instead of subjective expert assessment. The main objective of the research is to identify patterns in the relationship between time spent on social networks and self-reported indicators of mental well-being, including symptoms related to ADHD, anxiety, self-esteem, and depression. Data was collected through an anonymous survey administered via the LMS platform of SUAI. The sample included 473 participants, with 75% under the age of 35. Preprocessing steps involved cleaning outliers, imputing missing values, and formatting the data for analysis. Correlation matrices and heatmaps were created, followed by clustering using the k-means method. A stacked meta-model based on logistic regression and Gaussian Naive Bayes with a random forest as the final estimator was used for classification. The study revealed distinct user groups with varying levels of vulnerability to the influence of social media. The results can be used to develop intelligent systems for monitoring mental health risks and delivering personalized digital recommendations. The article is relevant to researchers in system analysis and applied machine learning.

Keywords: system analysis, digital activity, social networks, machine learning, clustering, correlation analysis, digital addiction, psycho-emotional state, information mining