Optimizing Hypothyroid Diagnosis with Physician-Supervised Feature Reduction using Machine Learning Techniques
Priyanshu Rawat, Madhvan Bajaj, Satvik Vats, Vikrant Sharma, Lisa Gopal, Rakesh Kumar
Abstract
Seven out of ten women in India are allegedly affected by thyroid difficulties, making hypothyroid diseases a major health concern among women. It is essential to detect hypothyroidism early in order to avoid problems and enhance patient outcomes. ML algorithms are a powerful tool for medical data analysis, and a number of researchers have employed different ML algorithms for hypothyroid categorization. In this study, we suggest the use of three supervised ML algorithms for hypothyroid classification: Random Forest, Logistic Regression, and Light-GBM Classifier, together with feature reduction. The Light-GBM Classifier achieves an accuracy of99.85% for 25 characteristics and 99.92% for 17 attributes, respectively. The inclusion of physician assistance to lower the number of variables has no significant effect on the accuracy of the Light-GBM classifier, indicating that fewer features may be necessary for reliable hypothyroid prediction. Our findings indicate that the Light-GBM classifier has the potential to be a valuable tool for hypothyroid prediction; however, additional validation and assessment on bigger and more varied datasets are required to assure its reliability and generalizability.