The thyroid gland is a vital component in controlling metabolic processes and is commonly known as the butterfly gland because of its form. It is situated in the neck. Thyroid lumps, cancer, hypothyroidism, and hyperthyroidism are among the disorders that might affect it. According to the latest research, forty-two million individuals in India suffer from thyroid diseases. For precise identification and successful therapy, these disorders must be identified promptly. Prior research has mostly used qualitative methods to examine one risk variable associated with the onset of an illness at some point. Yet, this method is ineffective and often overlooks the intricate relationships between variables, which leads to a great deal of debate among academics over the risk variables that have been discovered. Though there are still unanswered questions about the detection of subtypes and their cohabitation, the combination of deep learning (DL) approaches with healthcare scanning for computer-assisted diagnostic (CAD) platforms development has demonstrated potential in diagnosing the illness. More significantly, current CAD platforms are not very good at adjusting to various sample kinds. To overcome these obstacles, this study intends to boost diagnosis efficiency, generalize DL-oriented decision support systems (DSS), and provide insight into the pathophysiology of thyroid illness. A comparative evaluation of machine learning algorithms and DL algorithms is performed. According to the performance assessment, the convolution neural network and LSTM yield superior outcomes, with the maximum accuracy of 99.618% and 99.31%, respectively, and the lowest error percentages of 0.043 and 0.071. The structures suggested in this investigation have a significant influence on the community at large, further thyroid disease investigation, and improve medical care in the identification and treatment of linked conditions. © 2025 Elsevier B.V., All rights reserved.