Table 3.
Performance comparison among respective machine learning algorithms for predicting metabolic syndrome presence
| Training set | Test set | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
|
|
|
|||||||||
| AC | SP | SN | F1 | BCR | AC | SP | SN | F1 | BCR | |
| MLP | ||||||||||
| Model A | 95.75 | 1 | 1 | 0 | 0 | 81.16 | 1 | 1 | 0 | 0 |
| Model B | 97.81 | 0.99 | 0.515 | 0.672 | 0.717 | 78.90 | 0.93 | 0.12 | 0.167 | 0.333 |
|
| ||||||||||
| NB | ||||||||||
| Model A | 94.24 | 0.98 | 0.12 | 0.147 | 0.337 | 71.43 | 0.79 | 0.38 | 0.334 | 0.289 |
| Model B | 94.44 | 0.98 | 0.13 | 0.167 | 0.351 | 73.24 | 0.80 | 0.42 | 0.360 | 0.327 |
|
| ||||||||||
| RF | ||||||||||
| Model A | 98.78 | 1 | 0.71 | 0.832 | 0.844 | 78.80 | 0.95 | 0.08 | 0.122 | 0.272 |
| Model B | 99.71 | 1 | 0.93 | 0.966 | 0.967 | 82.14 | 0.99 | 0.01 | 0.014 | 0.083 |
|
| ||||||||||
| CT | ||||||||||
| Model A | 95.75 | 1 | 0 | 0 | 0 | 81.16 | 1 | 0 | 0 | 0 |
| Model B | 95.66 | 1 | 0 | 0 | 0 | 82.20 | 1 | 0 | 0 | 0 |
|
| ||||||||||
| SVM | ||||||||||
| Model Aa | 95.75 | 1 | 0 | 0 | 0 | 81.16 | 1 | 0 | 0 | 0 |
| Model Bb | 95.66 | 1 | 0 | 0 | 0 | 82.20 | 1 | 0 | 0 | 0 |
AC, accuracy; SP, specificity; SN, sensitivity; F1, F1 score; BCR, balanced classification rate; MLP, multilayer perceptron; NB, Naïve Bayes classification; RF, random forest classification; CT, decision tree classification; SVM, support vector machine classification; SNP, single nucleotide polymorphism.
Attributes for each model.
a
Model A: age, sex, body mass index, smoking, alcohol consumption, exercise
b
Model B: Model A + rs3764261, rs247617, rs2266788, rs964184, rs10830963, rs1260326, rs10830962, rs1883025, rs1919128, rs11757661 SNPs.