A neural network model for recognizing indicators of presuicidal risk in adolescents in the digital educational environment

Abstract

Relevance . The digitalization of education enables the application of machine learning methods and neural network models to monitoring and decision support in the educational environment. One of the most socially significant tasks is the early automated identification of adolescents’ presuicidal risk, the effectiveness of which critically depends on a well-grounded set of model input features. The aims of the study presented in the article are to develop the architecture of a neural network model for recognizing indicators of adolescents’ presuicidal risk in the digital educational environment and to empirically substantiate its set of input features based on self-attitude indicators and the riskogenity of the educational environment. Materials and methods. The empirical basis for feature selection consisted of data from 599 students aged 13–17 years (M = 14.34; SD = 0.83). The Self-Attitude Inventory by S. R. Pantileev, the Piers—Harris Self-Concept Scale, the inventory of the propensity for victim behavior, the personal differential and the “Index of sociocultural safety of a schoolchild” were used. Feature selection and validation were performed using correlation analysis, Student’s t-test and one-way analysis of variance (ANOVA); a fully connected neural network architecture with a three-class output was proposed. Results . A compact, empirically substantiated set of input features was formed: self-worth, self-acceptance, withdrawal, self-blame, behavior, family and the riskogenity of the educational environment. A high level of propensity for self-injurious and self-destructive behavior, corresponding to the target high-risk class, was found in 29 % of adolescents. Conclusions . The proposed neural network model with an empirically grounded feature set can be integrated into the digital educational environment as a tool for scalable early identification of presuicidal risk, subject to ethical constraints and confirmation by a specialist.