Table 5 Ranking of the 10 most important variables for algorithms run for predicting incident AF (among 27 clinical variables).
Ranking of variables | Traditional regression analysis | Support vector machines with linear Kernel | Decision tree | Random forest | Extreme gradient boosting |
|---|---|---|---|---|---|
1 | Heart failure | Heart failure | Age | Serum eGFR | Heart failure |
2 | Systolic blood pressure | Systolic blood pressure | Serum eGFR | Systolic blood pressure | Systolic blood pressure |
3 | Age | Age | Heart failure | Age | Age |
4 | Previous ischemic stroke/TIA | Previous ischemic stroke/TIA | Systolic blood pressure | Heart failure | PM2.5 |
5 | PM2.5 | PM2.5 | Previous ischemic stroke/TIA | PM2.5 | Serum triglyceride |
6 | Serum eGFR | Serum eGFR | PM2.5 | Sex | Serum total cholesterol |
7 | Serum triglyceride | Previous MI | Serum triglyceride | BMI | Serum HDL cholesterol |
8 | Sex | Sex | Sex | Smoking history | BMI |
9 | Smoking history | Smoking history | Smoking history | Fasting blood glucose | Serum eGFR |
10 | Serum total cholesterol | BMI | BMI | Previous ischemic stroke/TIA | Serum LDL cholesterol |
