Application Of Artificial Neural Network In The Research Of TCM Clinical Syndrome Differentiation Model

Based on the theory and experiment research of TCM syndrome differentiation model,this study adopts the multi sign classification algorithm based on artificial neural network to model the clinical data.Explore the big data machine learning,study the new method of TCM syndrome differentiation.ObjectiveIntelligent syndrome differentiation in traditional Chinese medicine needs massive computer calculations and repeating manual interventions.The study is to develop an auto-updating intelligent differentiation system to adapt the continuous renewal of traditional Chinese medicine theories and practices.MethodA total of 1146 cases were collected via traditional Chinese medicine syndrome element identification system.Cases were further divided randomly into training group(764 cases)and test group(384 cases).Symptoms and signs,syndrome elements and syndrome types of each selected case were recorded.Machine-learning was carried out by multi-label classification algorithm based on artificial neural network,and the trained model was tested in three aspects,namely syndrome type prediction by symptoms and signs,syndrome element prediction by symptoms and signs,and syndrome type prediction by syndrome element.Hidden neuron parameters and input/output activation function was adjusted and verified in each aspect to obtain a most optimized model.ResultsThe parameters results were average precision(0.79),coverage(8.18),one-error(0.18),hamming loss(0.94),ranking loss(0.05)for syndrome element prediction by symptoms and signs,and average precision(0.38),coverage(87.05),one-error(0.68),hamming loss(1.00),ranking loss(0.12)for syndrome type prediction by syndrome element,and average precision(0.25),coverage(98.48),one-error(0.84),hamming loss(1.00),ranking loss(0.15)for syndrome type prediction by symptoms and signs.Conclusion1.By implementing multi-label classification algorithm based on artificial neural network,machine-learning average precision achieved an exceeding result of 0.79 in syndrome element prediction by experimental syndrome under a condition of scare data.The improvement of machine-learning experimental results is about to supported by massive collected data,thus multi-label classification algorithm based on artificial neural network is a promising approach in machine-learning algorithm for traditional Chinese medicine syndrome differentiation.2.The results of syndrome type prediction by symptoms and signs,and that of syndrome type prediction by syndrome element,were unsatisfactory in present study.By analyzing experimental data,massive supportive data and balanced distribution showed a great significance to successful machine-learning as well as clinical practice.