KMAN Publication Inc. (KMANPUB) Journal of Personality and Psychosomatic Research (JPPR) 3041-8542 4 Serial Number 15 2026 07 01 Fairness-Aware Machine Learning for Predicting Diabetes Distress from Conscientiousness, Self-Efficacy, Medication Adherence, and Glycemic Control Fairness-Aware Machine Learning for Predicting Diabetes Distress from Conscientiousness, Self-Efficacy, Medication Adherence, and Glycemic Control 1 20 EN Journal Article 2026 03 17 <p>This study aimed to develop, compare, interpret, and assess the fairness of machine-learning models for predicting clinically meaningful diabetes distress from conscientiousness, diabetes management self-efficacy, medication adherence, and glycemic control among adults with type 2 diabetes in Kenya. This multicenter analytical cross-sectional study included 684 adults with type 2 diabetes recruited from four outpatient diabetes clinics in Kenya. Diabetes distress, conscientiousness, diabetes management self-efficacy, and medication adherence were assessed using standardized instruments, while glycated hemoglobin was extracted from clinical records. Clinically meaningful diabetes distress was defined as a mean Diabetes Distress Scale score of 2.0 or higher. Penalized logistic regression, support vector machine, random forest, and gradient-boosted decision-tree models were developed using a stratified 70:30 development-test split and repeated ten-fold cross-validation. Model performance was evaluated using discrimination, calibration, and classification metrics. Fairness was assessed across sex, age, and residence using differences in true-positive and false-positive rates, equalized odds, average odds, and selection-rate ratios. Shapley additive explanations were used to interpret predictor contributions. Participants with clinically meaningful distress had significantly lower conscientiousness, t(682) = 15.48, p &lt; .001, and self-efficacy, t(682) = 22.79, p &lt; .001, as well as significantly greater medication nonadherence, t(682) = −13.89, p &lt; .001, and higher HbA1c, t(682) = −11.11, p &lt; .001. Conventional gradient boosting achieved the highest AUROC of .842, sensitivity of .824, and specificity of .774. After fairness mitigation, AUROC remained .833 and sensitivity remained .824, while equalized-odds differences declined to .034 for sex, .018 for age, and .021 for residence. HbA1c had the greatest predictive contribution, followed by self-efficacy, medication nonadherence, and conscientiousness. Fairness-aware gradient boosting predicted diabetes distress with good accuracy while substantially reducing demographic disparities, indicating that combining psychological, behavioral, and metabolic variables may support equitable and timely psychosocial screening in diabetes care.</p> https://www.journals.kmanpub.com/index.php/jppr/article/download/5995/11501