A Thammasat team compared multiple machine-learning models for binary and three-class ophthalmology outpatient satisfaction. Nested cross-validation and training-only imbalance correction were followed by SHAP interpretation, with length of stay emerging as the strongest predictive feature.
Key findings
- Gradient Boosting with G-SMOTENC led the binary task, while Random Forest led the three-class task. Length of stay ranked highest, followed by demographic and visit variables, although the abstract omits numeric performance values.
Why this matters globally
Explainable AI may help hospitals prioritize service problems without relying solely on black-box predictions, and the validation workflow is reusable.
Thai researcher contribution
Thammasat University researchers developed the case study in a Thai university hospital, with Warut Pannakkong as corresponding author.
Limitations to consider
Single-clinic survey data may carry response bias. There is no causal identification or external validation, and the abstract does not report accuracy or calibration values.