Fig. 1: Comparison of feature importance between hyperthyroidism and hypothyroidism classification models. | Communications Medicine

Fig. 1: Comparison of feature importance between hyperthyroidism and hypothyroidism classification models.

From: Development and preliminary validation of a machine learning system for thyroid dysfunction diagnosis based on routine laboratory tests

Fig. 1

The blue line shows the feature importance of the hyperthyroidism classification model, while the red line shows that of the hypothyroidism classification model. a The five most important features in the hyperthyroidism model using Feature set 1 were S-Cr, MCV, total cholesterol, ALP, and albumin. The five most important features in the hypothyroidism model using Feature set 1 were S-Cr, LDH, total cholesterol, MCHC, and TP. b Among the five laboratory tests used as features, total cholesterol exhibited the highest feature importance in both the hyperthyroidism and hypothyroidism models. The second and third most important features were ALT and sex in the hyperthyroidism model, while those in the hypothyroidism model were AST and ALT. AST denotes aminotransferase, ALT: alanine aminotransferase, γ-GTP: gamma-glutamyl transpeptidase, RBC: red blood cell count, S-Cr: serum creatinine, ALP: alkaline phosphatase, UA: uric acid, LDH: lactic acid dehydr.ogenase, TP: total protein, BUN: blood urea nitrogen, A/G: albumin/globulin ratio, TB: total bilirubin, WBC: white blood count, Hb: hemoglobin, MCV: mean corpuscular volume, MCH: mean corpuscular hemoglobin, and MCHC: mean corpuscular hemoglobin concentration.

Back to article page