Predicting a few or many friends in schoolchildren: a machine learning approach

Area of Science:

• Child Psychology
• Machine Learning Applications
• Social Perception

Background:

• Understanding factors influencing children's social perceptions is crucial for developmental research.
• Previous studies highlight the impact of psychological variables on social relationships in youth.
• Sociodemographic and psychological factors may influence how children perceive their friendships.

Purpose of the Study:

• To investigate the predictive power of sociodemographic and psychological variables on friendship perception in Peruvian children aged 8-12.
• To identify key predictors of whether children believe they have many or few friends.
• To apply machine learning algorithms for analyzing friendship perception.

Main Methods:

• Utilized machine learning algorithms (Random Forest, XGBoost, SVM, Neural Network, Logistic Regression, Decision Tree) with stratified 10-fold cross-validation.
• Employed SMOTE (Synthetic Minority Over-sampling Technique) to address class imbalance in friendship perception.
• Assessed psychological variables including depression (PHQ-8), subjective well-being (SWB-3), sorrow, and bullying, alongside sociodemographic data like birthplace.

Main Results:

• Logistic Regression achieved the highest predictive performance (ROC-AUC 0.663 CV, 0.673 test set) with no overfitting.
• Subjective well-being emerged as the most significant predictor, followed by depression and birthplace.
• Optimizing the Youden's J index threshold improved classification accuracy from 64.6% to 72.8%.

Conclusions:

• Psychological factors, particularly subjective well-being, play a substantial role in shaping children's perception of their friendships.
• Machine learning models can effectively predict friendship perception based on psychological and sociodemographic data in children.
• Subjective well-being appears to be a more influential factor than clinical symptomatology in children's friendship perception.