Feature Selection And Syndrome Prediction For Rheumatoid Arthritis In Traditional Chinese Medicine

Traditional Chinese medicine(TCM)treatment is one of the safe and effective methods for Rheumatoid Arthritis(RA).TCM syndrome summarizes the origin and state of the disease and comprehensively reflects the degree of the disease.The types of syndrome are the best reflection on the real severity of diseases with overall consideration of all respects.Therefore,the accurate diagnosis of syndromes is critical to physicians before treatment and prescribing medicine.In our paper,several algorithms were introduced to predict TCM syndromes to help identify some important symptoms and physical indexes.A total of 1,713 records were collected from the First Affiliated Hospital of Anhui University of Chinese Medicine.Patients 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 all 200 features were extracted from an electronic medical record,including demographic,history and symptom information.Based on statistics and machine learning,we proposed 2 phase to predict RA syndromes.In the first phase,the critical features were filtered from the original features based on 5 rules.In the second phase,4 syndromes were predicted automatically based on 6 prediction algorithms.As a result,42 features were selected as critical from 200 features.The prediction accuracy algorithms combining feature selection was higher than when using all features.ANN have the maximum value of 0.88 in the subset of feature.Feature selection is proved to be efficient to improve prediction accuracy and ANN is a relative efficiency model applied to sparse and high-dimentional data in the field of TCM diagnosis.And we compare the prediction accuracy of different syndromes in ANN model.The accuracy was decrease with the sample size of syndromes.Feature selection methods and prediction technique were applied to mining the TCM syndrome type.Training model using selected features was more efficient than training model by all the features.ANN has the highest prediction accuracy among six machine learning algorithms.