Fig. 3 | Scientific Reports

Fig. 3

From: Development and external validation of an interpretable machine learning model for the prediction of intubation in the intensive care unit

Fig. 3

Model performance evaluation. Comparison between the full and simplified CatBoost models was conducted. The full model was developed using all available features, while the simplified model was derived based on key features selected by the recursive feature elimination algorithm. (a) AUC of the CatBoost model with varying numbers of variables was evaluated. (b) Feature importance in CatBoost with 5 variables was analyzed. (c) Receiver operating characteristic curves (ROCs) were compared between the full and simplified models. (d) Validation of our model’s performance in the eICU-CRD database involved different cohorts divided by ICU type. eICU represented all types of ICUs in the eICU-CRD database, Med-Surg referred to medicine-surgery ICUs, MICU indicated medicine ICUs, and SICU denoted surgery ICUs.

Back to article page