Authors: Palanichamy Jaganathan; Nallamuthu Rajkumar
Addresses: PG Department of Computer Applications, PSNA College of Engineering and Technology, Dindigul, Tamilnadu, 624 622, India. ' PG Department of Computer Applications, PSNA College of Engineering and Technology, Dindigul, Tamilnadu, 624 622, India
Abstract: Thyroid diseases have become common disorders worldwide. Thyroid is a gland in the neck that controls and stimulates the metabolism of the body. Hyperthyroidism (over production of thyroid hormone) and hypothyroidism (less production of thyroid hormone) are the most common problems of thyroid. In this study, improved F-score feature selection method is used to select the most relevant features for classification of thyroid dataset. The average improved F-score value of the features computed is set as the cut-off for selecting the features. The features above the cut-off are chosen as more relevant than others. Then the selected features were used in thyroid disease diagnosis with multilayer perceptron (MLP) and C4.5 algorithm and compared with recent research results obtained from generalised discriminant analysis-wavelet support vector machine (GDA-WSVM) technique (91.86%). The results show that our new feature selection method applied to this dataset has produced better classification accuracy than GDA-WSVM combination, with an improvement of 1.63% of accuracy for improved F-score-MLP combination (93.49%).
Keywords: expert systems; optimisation; thyroid diagnosis; multilayer perceptron; MLP; C4.5; thyroid diseases; hyperthyroidism; hypothyroidism; feature selection; intelligent diagnosis.
International Journal of Computational Science and Engineering, 2012 Vol.7 No.3, pp.232 - 238
Received: 16 Jul 2011
Accepted: 07 Feb 2012
Published online: 24 Jul 2012 *