ELMAN DEEP NEURAL NETWORK WITH MAXIMAL BOUNDARY PRINCIPAL FEATURE EXTRACTION FOR THYROID DISEASE DIAGNOSIS

Medical disease classification for thyroid using deep learning algorithms is a demanding chore, which can include insufficient, undetermined and vague information. The accessibility of such information in the datasets influences the classification performance. Currently, by evolution of deep learning algorithms, medical practitioners can minimize the number of probable flaws during the general diagnosis process and inspect the biomedical data in a timely and cost efficient manner. However, there has been minimum research works on the implementations of the machine learning classification algorithms in predicting thyroid disease. In this paper, an Elman Deep Neural Network with Optimized Principal Feature Extraction (EDNN-OPFE) is utilized for the analysis of thyroid medical data for disease classification. Initially, this method uses a Maximal Boundary Standard (MBS) factor Elman network to generate maximum margin separation classes that handle the over fitting problems. With this, the Maximal Boundary Optimized Principal Feature extraction model extracts only these features that are significantly contributing to the network. These reduced optimal features are again passed to the Elman Deep Neural Network model for classification according to the obtained results in to three different types, hyperthyroid, hypothyroid or normal. This proposed method is tested and validated through thyroid dataset and the performance is compared with the state-ofthe-art methods. This experimental analysis shows that our proposed method outperforms better as compared to other state-of-the-art methods in terms of disease diagnosing time, disease diagnosing complexity, accuracy and false positive rate.