Skip to main content

Table 1 Model performance comparison between sex-aggregated XGBoost, sex-separated XGBoost, semi-Bayesian ridge regression, Bozeman linear regression and linear regression with all possible interaction terms

From: Waist circumference prediction for epidemiological research using gradient boosted trees

Model:   XGBoost Semi-Bayesian Ridge Regression Bozeman Linear Regression Linear Regression
Sex:   Aggregated Separate Models Separate Models Separate Models Separate Models
  Count RMSE Bias RMSE Bias RMSE Bias RMSE Bias RMSE Bias
Overall 60,740 4.70 ± 0.05 0 ± 0.04 4.71 ± 0.04
0%
0 ± 0.05 4.89 ± 0.05***
4%
0 ± 0.04 5.01 ± 0.06***
7%
0 ± 0.04 4.72 ± 0.05
0%
0 ± 0.04
Female 26,750 5.41 ± 0.09 0.01 ± 0.07 5.43 ± 0.09
0%
0 ± 0.08 5.67 ± 0.1***
5%
0 ± 0.07 5.95 ± 0.12***
10%
0 ± 0.06 5.46 ± 0.1
1%
0 ± 0.06
 Asian 8402 4.4 ± 0.13 0.07 ± 0.17 4.39 ± 0.12
0%
− 0.01 ± 0.16 4.7 ± 0.12***
7%
0.68 ± 0.16*** 4.67 ± 0.12***
6%
0 ± 0.16 4.54 ± 0.11*
3%
0 ± 0.17
 Black 4321 5.94 ± 0.16 −0.06 ± 0.23 5.98 ± 0.17
1%
0.05 ± 0.2 6.24 ± 0.23**
5%
− 0.89 ± 0.23*** 6.67 ± 0.23***
12%
0.01 ± 0.29 5.91 ± 0.16
0%
0.01 ± 0.22
 Hispanic 5298 5.62 ± 0.31 −0.01 ± 0.36 5.65 ± 0.32
1%
− 0.01 ± 0.37 5.83 ± 0.34
4%
− 0.41 ± 0.36* 6.12 ± 0.31**
9%
0 ± 0.38 5.65 ± 0.31
1%
0 ± 0.35
 Other/Mixed 343 5.66 ± 0.68 − 0.83 ± 0.64 5.66 ± 0.56
0%
−0.55 ± 0.57 5.61 ± 0.79
− 1%
−0.91 ± 0.7 6.2 ± 1.15
10%
0.05 ± 0.67** 5.76 ± 0.66
2%
− 0.02 ± 0.8*
 White 8386 5.87 ± 0.16 0.04 ± 0.23 5.9 ± 0.15
0%
0 ± 0.24 6.12 ± 0.19**
4%
0.08 ± 0.25 6.53 ± 0.17***
11%
0 ± 0.27 5.9 ± 0.15
0%
0 ± 0.23
Male 33,990 4.05 ± 0.05 − 0.01 ± 0.07 4.05 ± 0.05
0%
0 ± 0.06 4.18 ± 0.05***
3%
0 ± 0.07 4.13 ± 0.05**
2%
0 ± 0.06 4.04 ± 0.05
0%
0 ± 0.06
 Asian 17,056 3.90 ± 0.08 − 0.02 ± 0.12 3.89 ± 0.08
0%
0.01 ± 0.11 3.98 ± 0.08*
2%
− 0.33 ± 0.11*** 4 ± 0.07 **
3%
0 ± 0.11 3.9 ± 0.08
0%
0 ± 0.12
 Black 3951 4.40 ± 0.24 0.13 ± 0.19 4.4 ± 0.23
0%
0.05 ± 0.19 4.8 ± 0.22**
9%
0.99 ± 0.2*** 4.56 ± 0.25
4%
0 ± 0.19 4.37 ± 0.24
− 1%
0 ± 0.19
 Hispanic 4691 3.88 ± 0.12 − 0.02 ± 0.14 3.88 ± 0.13
0%
− 0.02 ± 0.15 3.96 ± 0.12
2%
0.47 ± 0.16*** 3.9 ± 0.13
0%
0 ± 0.15 3.87 ± 0.12
0%
0 ± 0.15
 Other/ Mixed 381 4.52 ± 0.58 −0.25 ± 0.7 4.6 ± 0.62
2%
− 0.33 ± 0.63 4.75 ± 0.64
5%
0.67 ± 0.73* 4.7 ± 0.64
4%
1.2 ± 0.68*** 4.52 ± 0.51
0%
−0.03 ± 0.69
  White 7911 4.25 ± 0.12 − 0.05 ± 0.1 4.25 ± 0.12
0%
− 0.01 ± 0.09 4.37 ± 0.13*
3%
− 0.08 ± 0.09 4.28 ± 0.12
1%
−0.06 ± 0.09 4.23 ± 0.1
0%
0 ± 0.09
  1. Values represent mean ± standard deviation across the 10 iterations of each model
  2. Percentage change for RMSE is relative to sex-aggregated XGBoost model
  3. RMSE root mean squared error
  4. *p < .05, **p < .01 and ***p < .001 for statistical significance by two-tailed t-test versus sex-aggregated XGBoost model