Efficient Extraction and Automated Thyroid Prediction with an Optimized Gated Recurrent Unit in Recurrent Neural Networks

Abstract

Computer-aided tools are becoming increasingly important in medical diagnostics. This paper introduces the Efficient Feature Extraction Based Recurrent Neural Network (FERNN) for computer-aided thyroid disease prediction. The FERNN model uses a Gated Recurrent Unit Recurrent Neural Network (GRU-RNN) optimized with the COOT Optimization Algorithm.The study begins by gathering data from an open-source system and preprocessing it using min-max normalization to address missing values. The preprocessed data undergoes a two-level feature extraction (TLFE) procedure. In the first level, a ranked filter feature set technique is used to prioritize features based on medical expert recommendations. In the second level, a variety of metrics, including information gain, gain ratio, chi-square, and relief, are used to rank and select features. A composite measure guided by fuzzy logic is then used to select a judicious subset of features. The FERNN model uses the GRU-RNN to classify thyroid diseases in the databases. To optimise, the COOT optimization method is employed. The model's weights. The FERNN model was put into practise in MATLAB and assessed with a variety of statistical metrics, including kappa, accuracy, precision, recall, sensitivity, specificity, and the F-measure. The proposed methodology was benchmarked against traditional techniques, including the deep belief neural network (DBN), artificial neural network (ANN), and support vector machine (SVM).