An Efficient Selective Features Approach to Detect Hypothyroid Using Machine Learning

The thyroid gland is a hormone-producing gland located in the upper part of the neck. It helps to regulate the metabolism of the body. Its malfunction might lead to either inadequate or excessive thyroid hormone production. Thyroid issues are classified as hyperthyroidism or hypothyroidism. Predicting thyroid illness is now more crucial than ever. Machine learning algorithms and data mining techniques are critical for dealing with today’s massive amounts of data and information, particularly in the health-care system. A specific dataset is created that contains patient-related medical data. Based on the attributes considered, it is possible to categorize thyroid disease into three categories: primary hypothyroid, compensated hypothyroid, and routine. To identify earlier, various conventional machine learning methods are utilized, including Random Forest, Decision Tree, and Artificial Neural Network (ANN). These algorithms carry out the classification and determine the status but fall short of the desired results, particularly in the medical sector. In order to accomplish the goal, this research developed an effective classification strategy to improve the performance of hypothyroid prediction and it is divided into two stages: data preprocessing and classification methods. During the preparation phase, the dataset was preprocessed using two methods: PCA and an Extra Tree classifier, and the data was enriched with the Synthetic Minority Oversampling Technique (SMOTE). The proposed approach's outcomes were compared to traditional methods.