Calculating Diagnose Odd Ratio for Thyroid Patients using Different Data Mining Classifiers and Ensemble Techniques

Abstract

Thyroid hormones disorder is very common among people now-a-days. Various data mining techniques are used to identify thyroid problems. Machine learning provides help in multiple ways by different algorithms which are analyzed dataset and also generate different patterns. In this research paper we use three classifiers “Generalized Linear Model, Neural Network and Boosted Tree”. Two ensemble techniques Stacking and Random Forest are used to combine the results of three data mining classifiers. Each classifier provides different patterns of measuring the performance with thyroid dataset. The ensemble model behaves as a major classifier in which Random Forest is give more accuracy. All the experiment performed on three features of thyroid dataset Triiodothyronine (T3), Thyroxin (T4) and Thyroid Stimulating Hormone (TSH) different values. In this paper, we have proposed a new technique to find the pattern using Diagnose Odd Ratio (DOR) to find the patients that needs further treatment or not. Diagnose Odd Ratio (DOR) provides a value if this value is high then patient needs treatment and if its value is low patient does not need further treatment. Finally thyroid dataset has been analyzed and the Positive Likelihood Ratio (LR+), Negative Likelihood Ratio (LR-) and Diagnostic Odd Ratio (DOR) with the help of Sensitivity and Specificity are measure.