Fig. 3From: Assessing stroke severity using electronic health record data: a machine learning approachFeature selection based on cross-validated model performance. Feature selection based on cross-validated model performance on the training data. Mean model performance was calculated on 3-fold cross-validated subsets for feature selection. Maximum cross-validated performance was achieved with the top 226 features incorporated into the model; however, minimal gains in performance were seen beyond 100 featuresBack to article page