Predictive modelling of user satisfaction in the Oyo State Health Insurance Scheme using supervised machine learning techniques

The Oyo State Health Insurance Agency (OYSHIA) was established to enhance residents’ access to affordable and quality healthcare in Oyo State, Nigeria. This study assessed OYSHIA coverage and applied established machine learning approaches to identify predictors of users’ satisfaction and the most influential predictive features. A cross-sectional study was conducted among residents of Oyo State, Nigeria. The outcome variable was OYSHIA users’ satisfaction, with 13 explanatory variables covering sociodemographic characteristics, healthcare utilisation, and insurance-related factors. Class imbalance was addressed using class weighting during model training. Chi-square test of association was used to examine associations between explanatory variables and satisfaction. Multiple predictive models, including Logistic Regression (L2 and ElasticNet), Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), Support Vector Machine (SVM), and XGBoost were trained and evaluated. Within-sample performance was assessed using 10-fold stratified cross-validation and a test set, with mean area under the receiver operating characteristic curve (AUC), F1-score, precision, recall, and Cohen’s kappa used as evaluation metrics. A total of 190 residents participated in the study. Among the participants, 128 (67.4%) were enrolled in a health insurance scheme, of whom 118 (92.2%) were enrolled in OYSHIA. Distance to the hospital ( p < 0.001), scheme familiarity ( p = 0.013), and type of facility ( p = 0.004) were significantly associated with OYSHIA users’ satisfaction. The DT model achieved the highest mean test AUC across folds (0.836, 95% CI: 0.684–0.988), outperforming Logistic Regression models (L2: 0.660, 95% CI: 0.546–0.774; ElasticNet: 0.687, 95% CI: 0.478–0.896), RF (0.711, 95% CI: 0.451–0.971), GB (0.740, 95% CI: 0.594–0.886), SVM (0.658, 95% CI: 0.477–0.839) and XGBoost (0.760, 95% CI: 0.595–0.925). On the test set, the DT model achieved an F1-score, precision and recall of 0.647. Feature importance and explainability analyses identified type of healthcare facility used, distance to healthcare facility, and scheme familiarity as the most influential predictors of satisfaction. Established machine learning models identified key predictors of OYSHIA users’ satisfaction, particularly healthcare access, facility type, and scheme familiarity. These findings provide context-specific evidence to improve service delivery and support universal health coverage in Nigeria.