Identifying key influencers of patient satisfaction using an explainable machine learning approach

Patient satisfaction is a crucial measure of healthcare quality, influencing both health outcomes and care experiences. This study aims to identify the factors influencing patient satisfaction in healthcare facilities using machine learning algorithms due to their strong predictive capabilities. A cross-sectional survey was conducted with 312 patients from two private hospitals in Rangpur, Bangladesh. Machine learning models, including LightGBM, Random Forest, XGBoost, and CatBoost, were used to predict patient satisfaction, and SHAP value analysis was employed for interpretation. The LightGBM classifier (before SMOTE) outperforms other models across metrics, with the highest accuracy (0.85), MCC (0.69), and ROC-AUC(0.83) scores, establishing it as the optimal predictive model for this study. SHAP analysis further reveals that factors such as treatment plan, age, appointment ease, waiting time, and medication details significantly influence satisfaction levels. Overall, results indicate that structured interactions, shorter waiting times, and clear communication are associated with higher satisfaction, while extended waiting times and lack of decision involvement negatively impact patient experiences. To enhance patient satisfaction, healthcare providers should prioritize improving communication, reducing wait times, and offering clear treatment plans. Future research should explore additional factors to refine predictive models. The LightGBM classifier provides valuable insights into the key determinants of patient satisfaction, enabling healthcare practitioners and researchers to utilize it for targeted prediction and analysis in their specific contexts.