Machine learning prediction of psychological resilience: An analysis of health and self-perception variables

This study aims to evaluate the associations between psychological resilience and various health-related factors, including diet, smoking, alcohol use, and physical self-perception, in a sample of university students. The primary objective is to identify factors that are most significantly related to resilience and to determine their ability to predict an individual’s resilience level. A total of 360 university students participated in this cross-sectional study. Data were collected via validated questionnaires, such as the CD-RISC (Connor and Davidson, 2003) for psychological resilience, the PREDIMED (Guasch-Ferré, 2017) for the Mediterranean diet, the AUDIT (Kuitunen-Paul and Roerecke, 2018) for alcohol consumption risk, and the SCOFF (Morgan et al., 1999) for disordered eating behaviors. Additionally, questionnaires designed by researchers were used to assess smoking status and body image. Advanced machine learning models, specifically Random Forest (Breiman, 2001) ( Breiman, 2001 ) and TabNet (Arik and Pfister, 2021) were applied to predict the resilience levels. To address class imbalance, the SMOTE technique was used. The model performance was measured via the macro F1-score, a metric suitable for imbalanced datasets. Additionally, the importance values of the predictor variables were calculated to provide interpretability. A clustering analysis was also conducted to segment the population on the basis of their characteristics. Our predictive models achieved moderate accuracy but successfully identified the most influential variables for resilience. Perceived health and body image emerged as key predictors of psychological resilience. A positive association was also found between adherence to a healthy diet and increased resilience. Clustering analysis revealed two distinct resilience groups, with the lower resilience group showing a greater risk of alcohol dependence. Finally, our findings confirm the associations between psychological resilience and various health habits and self-perceptions. The study provides a novel applications of machine learning to a psychological construct, identifying key predictors and demonstrating the complexity of these relationships. While this research provides valuable insights, future studies should consider larger sample sizes and a longitudinal design to further explore these dynamics and incorporate a broader range of sociocultural and familial factors. • This study aims to evaluate the psychological factors, health habits, physical self-perception, and overall health of a sample of university students, as well as their impact on psychological resilience. • Methods: the study variables were collected using validated questionnaires and instruments specifically designed by the researchers. • The study also included a descriptive analysis and a statistical analysis designed to identify key relationships among these variables. • Additionally, advanced Machine Learning techniques, such as Random Forest, and Deep Learning methods have been applied. • Finally, a population segmentation was conducted through clustering analysis, enabling a deeper understanding of distinct subgroups within the data, which can provide insights for targeted interventions and personalized approaches.