Table 1.
The performance of models trained on the three populations separately and the combined populations. DNNs were trained and tested on four datasets: the South Korean, Canadian, Eastern European, and the combined dataset (All) with 19, 20, and 21 input features.
| Testing Set | Training Set | Features | r (all [f; m]) | R 2 (all [f;m]) | MAE (years) (All [f; m]) |
ε-Accuracy | Std of Δ Between Predicted and Actual Age |
|---|---|---|---|---|---|---|---|
| Canada | S. Korea | 20 | 0.57 [0.55; 0.59] | 0.24 [0.24; 0.24] | 7.87 [7.66; 8.12] | 0.70 [0.73; 0.69] | 6.19 [6.66; 5.69] |
| Canada | 20 | 0.70 [0.70; 0.70] | 0.52 [0.52; 0.47] | 6.36 [6.46; 6.28] | 0.80 [0.79; 0.81] | 5.24 [5.43; 5.05] | |
| 19 (wout sex) | 0.69 [0.69; 0.69] | 0.47 [0.47; 0.46] | 6.48 [6.51; 6.44] | 0.81 [0.79; 0.82] | 5.35 [5.55; 5.12] | ||
| E. Europe | 20 | 0.52 [0.52; 0.52] | 0.24 [0.27; 0.22] | 9.68 [9.98; 9.2] | 0.62 [0.58; 0.70] | 7.83 [8.34; 7.27] | |
| All | 21 | 0.72 [0.73; 0.71] | 0.52 [0.53; 0.50] | 6.17 [6.19; 6.17] | 0.70 [0.69; 0.72] | 5.12 [5.2; 5.02] | |
| 21 eq pop. Densities | 0.72 [0.73; 0.71] | 0.51 [0.52; 0.50] | 5.98 [6.02; 5.97] | 0.83 [0.82; 0.84] | 5.01 [5.09; 4.96] | ||
| 20 (wout pop.) | 0.72 [0.72; 0.71] | 0.51 [0.52; 0.50] | 6.05 [6.08; 6.03] | 0.83 [0.82; 0.84] | 5.03 [5.09; 5.01] | ||
| 20 (wout sex) | 0.69 [0.69; 0.68] | 0.44 [0.45; 0.44] | 6.62 [6.65; 6.59] | 0.80 [0.80; 0.81] | 5.21 [5.3; 5.06] | ||
| S. Korea | S. Korea | 20 | 0.70 [0.73; 0.66] | 0.49 [0.53; 0.43] | 5.59 [5.45; 5.77] | 0.85 [0.86, 0.84] | 4.76 [4.76; 4.76] |
| 20 (wout sex) | 0.69 [0.72; 0.66] | 0.48 [0.52; 0.44] | 5.64 [5.41; 5.81] | 0.87 [0.88; 0.86] | 4.8 [4.81; 4.8] | ||
| Canada | 20 | 0.52 [0.52; 0,47] | 0.24 [0.27; 0.22] | 7.1 [7.66; 8.12] | 0.72 [0.72; 0.71] | 6.32 [6.52; 6.03] | |
| E. Europe | 20 | 0.54 [0.56; 0.52] | 0.29 [0.27; 0.31] | 9.77 [10.21; 9.02] | 0.65 [0.62; 0.68] | 7.34 [7.49; 7.1] | |
| All | 21 | 0.70 [0.73; 0.66] | 0.49 [0.53; 0.43] | 5.60 [5.46; 5.78] | 0.75 [0.74; 0.76] | 4.82 [4.82; 4.8] | |
| 21 eq pop. Densities | 0.67 [0.7; 0.64] | 0.45 [0.49; 0.4] | 6.49 [6.4; 6.57] | 0.81 [0.8; 0.82] | 5.19 [5.20; 5.19] | ||
| 20 (wout pop.) | 0.69 [0.72; 0.65] | 0.46 [0.49; 0.41] | 5.78 [5.51; 5.91] | 0.86 [0.86; 0.86] | 4.85 [4.84; 4.86] | ||
| 20 (wout sex) | 0.68 [0.72; 0.64] | 0.46 [0.49; 0.41] | 5.77 [5.5; 5.91] | 0.86 [0.85; 0.86] | 4.85 [4.85; 4.85] | ||
| E. Europe | Korea | 20 | 0.76 [0.77; 0.75] | 0.34 [0.36; 0.33] | 8.52 [8.89; 7.93] | 0.31 [0.28; 0.35] | 6.47 [6.73; 5.96] |
| Canada | 20 | 0.68 [0.68; 0.72] | 0.27 [0.24; 0.27] | 9.25 [10.09; 7.91] | 0.38 [0.37; 0.38] | 7.28 [7.78; 6.17] | |
| E. Europe | 20 | 0.84 [0.85; 0.82] | 0.69 [0.72; 0.67] | 6.25 [6.24; 6.28] | 0.82 [0.83; 0.80] | 5.34 [5.51; 5.04] | |
| 19 (wout sex) | 0.83 [0.82; 0.84] | 0.70 [0.73; 0.68] | 6.29 [6.28; 6.31] | 0.83 [0.84; 0.80] | 5.39 [5.61; 5.12] | ||
| All | 21 | 0.84 [0.85; 0.82] | 0.70 [0.72; 0.67] | 6.22 [6.15; 6.33] | 0.81 [0.81; 0.81] | 5.28 [5.4; 5] | |
| 21 eq pop. Densities | 0.75 [0.75; 0.74] | 0.56 [0.57; 0.56] | 5.97 [5.94; 6.02] | 0.84 [0.84; 0.84] | 5.07 [5.29; 4.88] | ||
| 20 (wout pop.) | 0.82 [0.82; 0.82] | 0.68 [0.69; 0.68] | 6.57 [6.55; 6.61] | 0.81 [0.81; 0.81] | 5.37 [5.51; 5.09] | ||
| 20 (wout sex) | 0.82 [0.83; 0.82] | 0.67 [0.68; 0.67] | 6.29 [6.27; 6.32] | 0.81 [0.81; 0.81] | 5.3 [5.45; 5.06] | ||
| All | S. Korea | 20 | 0.73 [0.72; 0.73] | 0.44 [0.52; 0.53] | 7.01 [7.16; 6.82] | 0.85 [0.86; 0.84] | 5.85 [6.13; 5.46] |
| Canada | 20 | 0.65 [0.64; 0.66] | 0.36 [0.37; 0.33] | 8.71 [8.25; 7.54] | 0.88 [0.88; 0.88] | 6.49 [6.93; 5.87] | |
| E. Europe | 20 | 0.68 [0.69; 0.66] | 0.46 [0.48; 0.44] | 8.37 [8.78; 8.31] | 0.82 [0.83; 0.80] | 7.08 [7.31; 6.77] | |
| All | 21 | 0.80 [0.82; 0.78] | 0.65 [0.67; 0.61] | 5.94 [5.86; 6.04] | 0.83 [0.84; 0.82] | 5.04 [5.12; 4.93] | |
| 21 eq pop. Densities | 0.71 [0.72; 0.70] | 0.51 [0.53; 0.49] | 6.13 [6.05; 6.19] | 0.83 [0.84; 0.82] | 5.19 [5.27; 5.01] | ||
| 20 (wout pop.) | 0.79 [0.80; 0.78] | 0.62 [0.63; 0.61] | 6.09 [6.01; 6.14] | 0.83 [0.84; 0.82] | 5.15 [5.25; 5.01] | ||
| 20 (wout sex) | 0.78 [0.79; 0.78] | 0.61 [0.61; 0.61] | 6.23 [6.21; 6.25] | 0.82 [0.83; 0.81] | 5.32 [5.36; 5.11] | ||
| NHANES | S. Korea | 20 | 0.60 [0.57; 0.68] | 0.36 [0.32; 0.46] | 11.34 [12.53; 9.78] | 0.50 [0.46; 0.56] | 9.89 [10.27; 9.44] |
| Canada | 20 | 0.56 [0.57; 0.59] | 0.31 [0.32; 0.35] | 12.65 [13.84; 11.09] | 0.46 [0.41; 0.51] | 10.93 [10.96; 10.88] | |
| E. Europe | 20 | 0.68 [0.67; 0.70] | 0.46 [0.45; 0.49] | 7.88 [7.23; 8.37] | 0.40 [0.39; 0.42] | 10.00 [10.27; 9.44] | |
| All | 21 | 0.71 [0.71; 0.72] | 0.50 [0.50; 0.52] | 9.93 [10.36; 9.35] | 0.58 [0.57; 0.61] | 10.57 [11.00; 10.09] |
Note: Models trained on the combined dataset more accurately predict age of individual population samples compared to the population specific models. Exclusion of both sex and population features from training datasets resulted in a decrease in the accuracy of the age prediction. r for Pearson correlation coefficient; R2 for coefficient of determination; MAE for mean absolute error, that shows the average disagreement between actual chronological and predicted ages; f for female and m for male; “wout sex” stands for sets without sex as a feature and “wout pop.” stands for sets without population as a feature; ε-accuracy the accuracy of prediction within a period, which was calculated for ε of 10 years; Std for standard deviation; Δ for delta. The performance of best models (in terms of R2) is in bold.