Rapeutic Intervention Melitracen Biological Activity scoring Method; SNAPPE-II: Score for Neonatal Acute Physiology Perinatal Extension II; AUC: location beneath the curve, 95 CI: 95 confidence interval; compared with NTISS score; # compared with SNAPPE-II score.Figure 2. Comparisons of neonatal intensive unit mortality prediction models for instance as random forest, NTISS, Figure 2. Comparisons of neonatal intensive carecare unit mortality prediction models suchrandom forest, NTISS, and and SNAPPE-II within the set. (A) (A) Receiver operating characteristic curves of all machine studying models, the NTISS, the SNAPPE-II inside the test test set. Receiver operating characteristic curves of all machine learning models, the NTISS, and as well as the SNAPPE-II. (B) Decision curve analysis of all machine learning models, the NTISS, as well as the SNAPPE-II. bagged CART: SNAPPE-II. (B) Decision curve evaluation of all machine studying models, the NTISS, as well as the SNAPPE-II. Bagged CART: bagged classification and regression tree; NTISS: Neonatal Therapeutic Intervention Scoring Program; SNAPPE-II: Score bagged classification and regression tree; NTISS: Neonatal Therapeutic Intervention Scoring Program; SNAPPE-II: Score for for Neonatal Acute Physiology Perinatal Extension II. Neonatal Acute Physiology Perinatal Extension II.Amongst the machine learning models, the performances in the RF, bagged CART, and Amongst the machine understanding models, the performances with the RF, bagged CART, and SVM models were significantly greater than these from the XGB, ANN, and KNN models SVM models had been drastically improved than these in the XGB, ANN, and KNN models (Supplementary Materials, Table The RF RF bagged CART models also had signifi(Supplementary Materials, Table S2). S2). The andand bagged CART models also had considerably greater accuracy F1 F1 scores than XGB, ANN, and KNN models. In In addition, cantly higher accuracy andand scores than the the XGB, ANN, and KNN models.addition, the the model has features a substantially much better AUC value than the bagged CART model. RF RF model a considerably much better AUC value than the bagged CART model. TheThe calibration belts ofRF and bagged CART models and the standard scoring calibration belts from the the RF and bagged CART models along with the traditional scoring systems for NICU mortality prediction are Figure 3. The RF model showed much better systems for NICU mortality prediction are shown inshown in Figure three. The RF model showed far better calibration among neonates with respiratory failure whoa highat a high threat of morcalibration among neonates with respiratory failure who had been at were risk of mortality tality the NTISS and SNAPPE-II scores, specially when the predicted values have been than did than did the NTISS and SNAPPE-II scores, especially when the predicted values had been larger than greater than 0.8.83. 0.8.83.Biomedicines 2021, 9, x FOR PEER Assessment Biomedicines 2021, 9,8 7of 14 ofFigure 3. Calibration belts of (A) random forest, (B) bagged classification and regression tree Figure 3. Calibration belts of (A) random forest, (B) bagged classification and regression tree (bagged CART), CART), (C) NTISS, SNAPPE-II for NICU mortality prediction within the test the (bagged (C) NTISS, and (D) and (D) SNAPPE-II for NICU mortality prediction inset. test set.3.two. Rank of Predictors within the Prediction Model three.two. Rank of Predictors inside the Prediction Model A total of 41 variables or features had been employed to create the prediction model. Of A total of 41 variables or options were utilised to develop the prediction m.
