Predicting Thyroid Disease using Optimized Multi-Layered Long Short-Term Memory (LSTM) Neural Network with Residual Attention Mechanism

Abstract - Thyroid disease is a common ailment that arises when the thyroid gland produces an inadequate amount of thyroid hormone. The challenge in diagnosing thyroid disease is that its symptoms resemble those of other conditions, making it difficult to detect. One of the most effective ways to diagnose thyroid problems is through blood tests, which measure the amount of thyroid hormones in the bloodstream. However, interpreting the complex data generated by these tests can be challenging. Early detection of thyroid issues is crucial to prevent complications and reduce mortality rates. To improve decision-making systems, medical data mining is increasingly using deep learning models. This study proposes a novel stacked residual long-short memory architecture (SR-LSTM) with an attention mechanism for predicting thyroid disease. The model employs an attention mechanism and layer stacking to enhance the accuracy of prediction. Additionally, this work is optimized using the Giant Trevally Optimizer (GTO) metaheuristic optimization algorithm. The performance results of the
proposed model achieve outstanding performance in precision, recall, F1, and accuracy, with scores of 99.836%, 99.839%,
99.837%, and 99.931% than traditional methods.