Feature selection and syndrome classification for rheumatoid arthritis patients with Traditional Chinese Medicine treatment

Abstract

IntroductionThe classification of TCM syndromes is central to understanding the nature of diseases and improving treatment. This study focuses on selecting critical features of demographic information, personal medical history and symptoms and improving the accuracy of syndrome classification. MethodsA total of 1713 records were collected from the First Affiliated Hospital of Anhui Chinese Medicine University. Five rules for feature selection and six models were applied to classify TCM syndromes. ResultsPatients with rheumatoid arthritis were diagnosed with one of four TCM syndromes: damp-heat obstruction syndrome (DHO, 60.5 %), phlegm and blood stagnation syndrome (PBS, 19.8 %), liver and kidney deficiency syndrome (LKD, 15.8 %), or wind-cold obstruction syndrome (WCO, 4 %). In total, 200 features were extracted from electronic medical records. From these, 42 were selected as critical features. The classification accuracy of using feature selection was higher than when using all features, with a maximum value of 0.88 for the Artificial neural network (ANN). ConclusionsFeature selection methods and classification techniques were applied to mine data on TCM syndromes. Feature selection improved the performance of the classification models. Of six algorithms, ANN had the highest accuracy for syndrome classification.