Improving Healthcare Outcomes: Doctor Classification and Performance Prediction via ML

Healthcare systems face challenges in objectively evaluating clinician performance and allocating resources efficiently. This study proposes a machine learning (ML) framework to classify doctors based on clinical outcomes, predict future performance, and enhance decision-making using the MIMIC-III database. We extracted structured (lab results, treatment histories) and unstructured (clinical notes) data to train four ML models: Random Forest, XGBoost, SVM, and Neural Networks. Random Forest achieved the highest classification accuracy (85%, AUC-ROC: 0.89), while XGBoost excelled in precision (82%) and recall (79%). Model interpretability was ensured using SHAP values, and rigorous validation (5-fold cross-validation, external eICU dataset) confirmed generalizability. Our results demonstrate that ML-driven clinician evaluation improves resource allocation, reduces diagnostic errors by 18%, and enhances patient satisfaction. This work bridges the gap between predictive analytics and healthcare management, offering actionable insights for hospitals to optimize care delivery.