Hybrid Machine Learning Framework for Predicting Family Adaptation to Chronic Illness

Abstract

Objective: The objective of this study was to develop and validate a hybrid machine learning framework for accurately predicting family adaptation to chronic illness using integrated psychosocial, relational, and clinical indicators.

Methods and Materials: This cross-sectional study recruited 812 family caregivers of patients with chronic illnesses from major healthcare centers in Chile using multistage cluster sampling. Data were collected through standardized measures assessing family functioning, caregiver burden, psychological distress, social support, and illness characteristics. A hybrid machine learning architecture integrating gradient boosting, random forest, support vector machine, and a meta-learning classifier was developed. Data were preprocessed using normalization, multiple imputation, and feature selection. The dataset was partitioned into training, validation, and testing subsets. Model training employed five-fold cross-validation with Bayesian hyperparameter optimization. Model interpretability was enhanced through explainable artificial intelligence techniques including SHAP analysis.

Findings: The hybrid model achieved high predictive performance on the independent test set, with an accuracy of 0.893, F1-score of 0.884, and AUC of 0.921, significantly outperforming individual machine learning models. Family functioning, caregiver burden, psychological distress, symptom severity, and social support emerged as the most influential predictors. The model demonstrated strong discriminative capacity between low, moderate, and high family adaptation levels and exhibited excellent calibration.

Conclusion: The proposed hybrid machine learning framework provides a robust, interpretable, and clinically valuable tool for early identification of families at risk of maladaptation to chronic illness, offering significant potential for improving family-centered care and psychosocial intervention planning.