Table 1.
Statistics on the raw dataset.
Table 2.
Features used for training the machine learning models.
Table 3.
Composition of the Feature-saving and Record-saving datasets.
Fig 1.
Pearson Matrix describing the relation among input features and the transition to the SP phase within 180 days.
Table 4.
Results of Visit-Oriented models on Feature-saving and Record-saving datasets at different time points.
Table 5.
Results of History-Oriented setting on Feature-saving and Record-saving datasets at 180, 360 and 720 days.
Fig 2.
The choice of confidence threshold has different effects on the indicated variables. Analysis was performed with Random Forest and LSTM for Record Preserving datasets.