Machine Learning-Based Prediction Model of Preterm Birth Using Electronic Health Record

Table 4

The top 20 importance variables of RF model.

Variables	Decreased accuracy
Age (physical examination)	0.0251
Magnesium (blood test)	0.0098
Fundal height (physical examination)	0.0077
Serum inorganic phosphorus (blood test)	0.0038
Mean platelet volume (blood test)	0.0038
Waist size (physical examination)	0.0038
Total cholesterol (blood test)	0.0035
Triglycerides (blood test)	0.0031
Globulins (blood test)	0.0024
Total bilirubin (blood test)	0.0024
Neutrophil granulocytes (blood test)	0.0024
Red blood cell distribution width-SD (blood test)	0.0024
Bacterial vaginosis (gynecological examination)	0.0021
Urine bilirubin (urine test strip)	0.0021
Urine white blood cell (urine test strip)	0.0021
Diastolic blood pressure (physical examination)	0.0014
Blood group (blood test)	0.0014
Parity (physical examination)	0.0014
Eosinophil granulocytes (blood test)	0.0010
White blood cell count (blood test)	0.0010

RF: Random Forest tree.