Using Machine Learning Techniques to Predict Adolescents’ Involvement in Family Conflict

Many cases of violence against children occur in homes and other close environments. Machine leaning is a novel approach that addresses important gaps in ways of examining this socially significant issue, illustrating innovative and emerging approaches for the use of computers from a psychological perspective. In this paper, we aim to use machine learning techniques to predict adolescents’ involvement in family conflict in a sample of adolescents living with their families (community adolescents) and adolescents living in residential care centers, who are temporarily separated from their families because of adverse family conditions. Participants were 251 Spanish adolescents (Mage = 15.59), of whom 167 lived in residential care and 84 lived with their families. We measured perceived interparental and family conflict, adolescents’ emotional security, emotional, cognitive, and behavioral immediate responses to analog interparental conflict (IPC), and adolescents’ sociodemographic variables (i.e., age, gender). With a prediction accuracy of 65%, our results show that adolescents in residential care are not at greater risk for involvement in family conflict compared to adolescents living with their families. Age and gender are not salient predictive variables. We could identify that responses to analog IPC, adolescents’ emotional security, triangulation in IPC, and the presence of insults or blame during family disputes predict adolescents’ involvement in family conflict. These results point to variables with a potential predictive capacity, which is relevant for research and intervention.

keywords: emotional security, simulated conflict,residential care, predictive technologies