Logistic Regression

This section describes experiments using a Logistic Regression classifier to predict whether a patient would, or would not, receive thrombolysis. Data is restricted to admissions within 4 hours of stroke onset (to units that have at least 300 patients, and 10 thrombolysis uses, over three years)

This section contains the following notebooks:

• Logistic Regression Classifier - Fitting to all stroke teams together: A Logistic Regression classifier that is fitted to all data together, with each stroke-team being a one-hot encoded feature. Analyses the models for: 1) Various accuracy scores, 2) Receiver-Operator Characteristic Curve, 3) Sensitivity-Specificity Curve, 4) model calibration, and 5) Learning rate.

• Logistic Regression Classifier - Fitting hospital-specific models: A Logistic Regression classifier that has a fitted model for each hospital. * Logistic Regression Classifier - Fitting to all stroke teams together: A Logistic Regression classifier that is fitted to all data together, with each stroke-team being a one-hot encoded feature. Analyses the models for: 1) Various accuracy scores, 2) Receiver-Operator Characteristic Curve, 3) Sensitivity-Specificity Curve, and 4) model calibration.