Decoding Physician Burnout: Leveraging Machine Learning for Early Detection on Survey-Based Features

This project aimed to predict physician burnout levels using machine learning techniques applied to self-reported survey data from the Montgomery County Medical Society. The hypothesis was that classification algorithms in machine learning can be used to help predict cases of physician burnout by finding relationships between variables, since burnout is a quantifiable factor. This project examined physician studies specifically, not healthcare professionals in general, including primary practice settings and specialties as factors that affect burnout levels. After data preprocessing, visualizations were made based on correlations between burnout levels and factors such as time spent on various tasks and years of experience. After accuracy-guided hyperparameter tuning on various models, a Random Forest Classifier model was used and evaluated with precision, recall, and F1-score metrics. The model performed best in predicting moderate burnout (with 62% accuracy for that class), but tended to misclassify low and high burnout levels as moderate. Figures provide visualizations of key relationships and classification results. These findings highlight both the potential and the limitations of using survey-based features for burnout prediction, given the variability and subjectivity of self-reported measures.