An Optimized Deep Ensemble Super-Learner Model For Thyroid Disease Classification

Abstract

Early detection of thyroid diseases is crucial to mitigate severe consequences. various Artificial Intelligence (AI) models have been developed to identify and classify thyroid diseases. But, certain models that rely on a single classifier often suffer from high prediction errors and lower generalizability and accuracy. To solve this, Optimized Deep Ensemble Super-Learner Model (ODESLM) is developed for thyroid disease prediction. In this method, Deep Ensemble Super-Learner Model (DESLM) is designed for feature selection tasks with six base learners like Support Vector Machine (SVM), Extreme Gradient Boosting (XGB), Random Forest (RF), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) and Bidirectional Gated Recurrent Unit (Bi-GRU).The Super Learner Model (SLM) optimally combines and weights the predictions of base models to minimize the overall prediction error. The ensemble method uses resampling or reweighting to build a stacked model to capture capturing diverse patterns for enhancing the accuracy of thyroid disease identification. The Output values of DESLM is optimized using Secretary Bird Optimization Algorithm (SBOA) which is inspired by secretary Birds' foraging behavior to avoid incorrect classifier selection in training datasets. SBOA refines training datasets, optimizes hyperparameters and identifies relevant features from base models to enhance the super-learner's performance. It adjusts model weights and parameters to balance bias and variance enhancing generalization and reducing overfitting. This optimized ensemble model improves accuracy on larger datasets with lowers predictive error in thyroid disease prediction.Experimental results show that proposed model achieves 92.73% accuracy on the thyroid disease dataset, outperforming traditional models.