2.6. Data analysis
We used MySQL and Tableau (version 2019.1.0) to extract data from MIMIC-III v1.4, and Python3.7 to build models and process data. The working environment was PyCharm 2018.3.2.
Each of machine learning models involves a parameter-tuning process, so we used the learning curve and grid search to select the parameters to get the best model (Table A.1 in Appendices). In this study, accuracy rate, precision rate (P), recall rate (R), F1 score, and area under ROC curve (AUROC) were used to measure the model performance.