LightGBM Analysis of Loneliness in Gifted Adolescents: Predictive Contributions of Maladaptive Perfectionism, Emotional Awareness, and Peer Support

Abstract

The present study aimed to employ a Light Gradient Boosting Machine (LightGBM) approach to predict loneliness among gifted adolescents based on maladaptive perfectionism, emotional awareness, and peer support and to determine the relative contribution of these variables to loneliness outcomes. This cross-sectional predictive study was conducted among 512 gifted adolescents aged 13 to 18 years recruited from specialized educational programs and secondary schools in Mexico. Participants completed standardized measures of loneliness, maladaptive perfectionism, emotional awareness, and peer support. Data were analyzed using Python and R statistical environments. After preliminary data screening and standardization procedures, a LightGBM regression model was developed using an 80/20 training–testing split. Hyperparameter optimization was performed through five-fold cross-validation and Bayesian optimization. Model performance was evaluated using the coefficient of determination (R²), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Feature importance analysis and SHapley Additive exPlanations (SHAP) were used to identify and interpret the contribution of predictor variables. The optimized LightGBM model demonstrated strong predictive performance, achieving an R² of .811 in the testing dataset, indicating that the model explained 81.1% of the variance in loneliness scores. Correlational analyses revealed that loneliness was positively associated with maladaptive perfectionism and negatively associated with emotional awareness and peer support. Feature importance analysis identified peer support as the strongest predictor of loneliness (42.6% relative importance), followed by maladaptive perfectionism (33.8%) and emotional awareness (23.6%). SHAP analyses confirmed these findings, showing that lower peer support and higher maladaptive perfectionism substantially increased predicted loneliness levels, whereas higher emotional awareness reduced loneliness risk. The consistency between LightGBM feature importance rankings and SHAP explanations supported the robustness and interpretability of the predictive model. The findings indicate that loneliness among gifted adolescents is shaped by a combination of interpersonal and intrapersonal factors, with peer support emerging as the most influential protective factor, maladaptive perfectionism functioning as a significant risk factor, and emotional awareness contributing additional resilience. The study demonstrates the utility of explainable machine learning approaches for identifying key determinants of psychological well-being in gifted populations and provides evidence to support interventions targeting social connectedness, perfectionistic tendencies, and emotional competencies.