A Traditional Chinese Medicine Syndrome Classification Model based on Cross-FGCNN: Model Development and Validation (Preprint)

UNSTRUCTURED Syndrome differentiation and treatment is the core of traditional Chinese medicine(TCM) in the treatment of diseases. TCM doctors can dialectically classify the syndrome according to the patients' symptoms and conduct treatment. Syndrome differentiation can be regarded as a mathematical model for multi-classification of different high-dimensional sparse symptom vectors. The FGCNN can quickly and effectively extract the nonlinear cross features of sparse vectors in the CTR task. On this basis, we selected the data of 5273 real cases of dysmenorrhea and divided the symptoms into field according to the four diagnosis of TCM, so as to construct an improved Cross-FGCNN model and apply it to the intelligent dialectics of TCM. We used 6 kinds of intelligent dialectical models and 3 kinds of CTR models as comparisons at the same time. Cross-FGCNN can achieve 96.21% accuracy and 0.836 Log-Loss, which is better than other models. We maintain that the model of Cross-FGCNN can automatically extract the linear and nonlinear features of symptoms and classify them, having great potential in the intelligent dialectics of TCM in the future.