A novel hybrid decision support system for thyroid disease forecasting

Abstract

Diagnosis of thyroid disease requires proper interpretation of functional data of the thyroid gland, which produces hormones to regulate the metabolism of human body. The thyroid disorders are classified on the basis of quantity of hormones produced, i.e., hyperthyroidism the case in which more hormones are produced and hypothyroidism where less than the required number of hormones are produced. Thyroid disease is a critical issue in underdeveloped countries, due to lack of awareness and early diagnosis. The use of machine learning methods is increasing with the passage of time as an alternative approach for the early diagnosis of thyroid disease. In this article, we present a novel intelligent hybrid decision support system based on linear discriminant analysis (LDA), k nearest-neighbor (kNN) weighed preprocessing, and adaptive neurofuzzy inference system (ANFIS) for the diagnosis of thyroid disorders. In the first stage of the LDA–kNN–ANFIS technique, LDA reduces the dimensionality of the disease dataset and eliminates unnecessary features. In the second stage, selected attributes are preprocessed using kNN-based weighed preprocessor. In the last stage, preprocessed selected attributes are provided to adaptive neurofuzzy inference system as an input for diagnosis. The proposed approach experimented on thyroid disease dataset, retrieved from the University of California Irvin’s machine learning repository to validate the overall performance of the system. The computed classification analysis made by accuracy, sensitivity, and specificity values of this approach were 98.5, 94.7, and 99.7%, respectively. This approach can also be efficiently applied to the diagnosis of other deadly diseases to get maximum accuracy with minimum possible features of the dataset.