Figure 6:

ROC curves for asthma exacerbation prediction, comparing (a) L₁-regularized logistic regression models, with and without the inclusion of large-vocabulary EHR categories (diagnoses and interventions), and (b) the best-performing logistic regression model (L₁ regularization and a temporal window-based representation), random forests (temporal window-based and last occurrence-based representations), and LSTM (sequence-based representation).