Table 4.
(A) Feature importance ranking with mean decrease Gini index calculated by the random forest (RF) algorithm for AA and (B) NHW.
| Feature selection method | Features selected | Classification method | Classification accuracy (%) |
|---|---|---|---|
| (A) | |||
| Boruta test—AA | 2-Oxoisocaproic acid, arginine, a-tocopherol, citric acid, histidine, maltose, methionine, n-acetylglutamic acid, o-phosphoethanolamine, oxalic acid | Random forest | 80.65 |
| Boruta test—AA | 2-Oxoisocaproic acid, arginine, a-tocopherol, citric acid, histidine, maltose, methionine, n-acetylglutamic acid, o-phosphoethanolamine, oxalic acid | Decision trees | 79.03 |
| Boruta test—AA | 2-Oxoisocaproic acid, arginine, a-tocopherol, citric acid, histidine, maltose, methionine, n-acetylglutamic acid, o-phosphoethanolamine, oxalic acid | Logistic regression | 77.42 |
| Boruta test—AA | 2-Oxoisocaproic acid, arginine, a-tocopherol, citric acid, histidine, maltose, methionine, n-acetylglutamic acid, o-phosphoethanolamine, oxalic acid | SVM | 74.19 |
| Recursive feature elimination—AA | Arginine, maltose, methionine, n-acetylglutamic acid, o-phosphoethanolamine | Random forest | 79.03 |
| Recursive feature elimination—AA | All features gave more accuracy than selecting some features | Decision trees | 66.04 |
| recursive feature elimination—AA | Arginine, maltose, methionine, n-acetylglutamic acid, o-Phosphoethanolamine | Logistic regression | 74.19 |
| Recursive feature elimination—AA | 2-Oxoisocaproic acid, arginine, histidine, maltose, methionine, n-acetylglutamic acid, o-phosphoethanolamine, oxalic acid | SVM | 75.81 |
| (B) | |||
| Boruta test—NHW | 3-Hydroxybutanoic acid, 9,12-octadecadienoic acid, 9-hexadecenoic acid, a-ketoglutaric acid, b-alanine, cholesterol, citric acid, lactamide, oxalic acid, palmitic acid, p-cresol, pyruvic acid, tetradecanoic acid | Logistic regression | 79.03 |
| Boruta test—NHW | 3-Hydroxybutanoic acid, 9-hexadecenoic acid, oxalic acid, palmitic acid, tetradecanoic acid | Random forest | 75.61 |
| Boruta test—NHW | 3-Hydroxybutanoic acid, 9-hexadecenoic acid, cholesterol, oxalic acid, palmitic acid, tetradecanoic acid | Decision trees | 60 |
| Boruta test—NHW | 3-Hydroxybutanoic acid, 9-hexadecenoic acid, cholesterol, oxalic acid, palmitic acid, tetradecanoic acid | SVM | 61.73 |
| Recursive feature elimination—NHW | 3-Hydroxybutanoic acid, 9-hexadecenoic acid, oxalic acid, palmitic acid, tetradecanoic acid | Logistic regression | 72.50 |
| Recursive feature elimination—NHW | 3-Hydroxybutanoic acid, 9-hexadecenoic acid, oxalic acid, palmitic acid, tetradecanoic acid | Random forest | 75.61 |
| Recursive feature elimination—NHW | 3-Hydroxybutanoic acid, 9-hexadecenoic acid, oxalic acid, palmitic acid, tetradecanoic acid | Decision trees | 60 |
| Recursive feature elimination—NHW | 3-Hydroxybutanoic acid, 9-hexadecenoic acid, oxalic acid, palmitic acid, tetradecanoic acid | SVM | 62.86 |
The top features selected are shown.