Ensemble-based Effective Diagnosis of Thyroid Disorder with Various Feature Selection Techniques

Thyroid illness is characterized by the abnormal growth of thyroid tissue on the thyroid gland's periphery. Hyperthyroidism and hypothyroidism are the two most common types of thyroid disease, which occur when this gland generates an abnormally large amount of hormones. The goal of this project was to create an efficient homogenous system, an ensemble of ensembles in conjunction with various characteristics selection approaches for the improved detection of thyroid illness. The dataset utilized is thyroid 0387 data from the open-source KEEL repository. Following the necessary preprocessing steps, three types of attribute selection strategies, recursive feature elimination (RFE), select from model (SFM), and select k-best (SKB) were used with multiple machine learning (ML) based attributes estimators. The bagging and boosting-based classifiers were activated by the homogeneous ensembling, and the voting ensemble used both soft and hard voting to classify the results. The performance of the model has been evaluated by using recall, sensitivity, accuracy, Cohen kappa, etc. The highest accuracy of 99.27% with 97% precision and Fl-score, 98% recall was achieved at a relatively low computational cost in 33 ms prediction time by using RFE with logistic regression estimator. The experimental results show the utmost effectiveness of the proposed technique for improved thyroid disease diagnosis by outperforming the comparable benchmark models in its area.