Aim: The purpose of this study is to use machine learning techniques, Bagged CART, to classify hypothyroidism, which typically results from insufficient thyroid hormone synthesis in the body or seldom affects target tissues, and to identify potential risk factors.
Material and Methods: In this study, the open source data set obtained from the UCI database was used. The 10-fold cross-validation technique was used in the creation of the Bagged CART model from the Decision Tree Ensembles class to classify hypothyroid, and the performance criteria of this model were accuracy, balanced accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1-Score, G-mean and Matthews Correlation Coefficient ( MCC) was given. Then, the significance of the variable was calculated through the model created and possible risk factors for hypothyroidism were determined.
Results: The accuracy, balanced accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1-Score, G-mean and Matthews Correlation Coefficient (MCC) performance criteria for the model created for the classification of hypothyroidism were 99.9%, 99.2%, 98.3%, 100.0%, 100.0%, 99.9%, 99.2%, 99.9%, and 99.1%, respectively. According to the created XGBoost model, the three most important factors that could be associated with hypothyroidism were determined as TBG, TSH, T4U, TT4, age, FTI, Query hypothroid, on thyroxine, on antithyroid medication, thyroid surgery, sex, TBG measured, sick, T3 mesured, Query hyperthyroid, goitre.
Conclusion: In conclusion, considering the results of the machine learning model created in this study, the hypothyroidism classification performance was quite high and the significance of the variables and possible risk factors for hypothyroidism were determined. In the light of the findings, it is predicted that these risk factors may be useful in the clinic.
Key words: Bagged classification and regression trees (CART), Machine learning, hypothyroidism, classification.
|
