Improved Ensemble Classification Method of Thyroid Disease Based on Random Forest

Abstract

The thyroid disease has already been the second largest in the field of endocrine, and the classification of disease is the primary problem in clinical treatment. In computer-aided diagnosis (CAD), machine learning techniques have been widely used to assist the medical experts in decision making. This paper proposed a new method for thyroid disease classification based on random forest. Principal Component Analysis is used to preserve the variability in the data. Rotation Transformation can enlarge the discrepancy of the base classifiers and improve the accuracy of the ensemble classifier. Our method performs much better than Bagging, Random forest and AdaBoost, and can solve the accuracy-diversity dilemma. Experimental results show that the classification accuracy of this method can reach to 95.63% on the dataset from UCI machine learning repository. In order to verify the effectiveness of the method furthermore, this paper also chooses the real clinical medical data set. It is more complex than the UCI standard dataset in data quantity and dimension. Compared with other methods, the classification accuracy of our method reaches to 96.16%.