Intelligent Diagnosis of Hypothyroidism Using Hybrid Machine Learning with Optimized Feature Engineering and Parameter Tuning

Abstract

The thyroid gland is considered one of the vital organs in the human body. It secretes hormones crucial for maintaining metabolism. This disease creates several abnormalities in the human body. It has a myopic characteristic that has been unnoticeable for detection. For the detection of the thyroid, several complicated clinical tests are being performed. Sometimes, due to statistical human errors, it is not possible to detect the precise status of the disease. This work, a hybrid machine learning based diagnostic screening system has been proposed for the accurate identification of thyroid disease. For boosting the performance of the system, less correlated features are removed by using the Chi-Square test. The hybrid machine learning-based classifier is formed by stacking different basic algorithms such as Random Forests, SVM, and KNN. The resultant system is optimized by performing hyper-parameter tuning using Grid Search CV. The highest accuracy obtained by the system is 99.79%.