Liver Yang Forming Wind Syndrome Model Research Based On MARS

Multivariate Adaptive Regression Splines (MARS) is a self-adaptable method based on fitting, also a processing method specializing for high dimensional data that is nonlinear, non-parametric, and very flexible, of strong generalization ability. The method takes product spline as the basis function and sample data coordinates as an optional node vector value, its procedure choose basis functions and node vector according to minimum Generalized Cross-Validation (GCV). In this paper, we focus on studying high dimensional modeling, and are attempt to explore the high dimensional model established.Liver Yang Forming Wind Syndrome (LYFWS), as a syndrome of Traditional Chinese medicine (TCM), is one of the three subtypes of endogenous Liver Wind Syndrome. To investigate the pathophysiology basis of this syndrome, many people used a variety of methods, such as literature accumulated, expertise, mathematical statistical analysis, to do a lot of research, but its essential connotation remains unclear. LYFWS has various characteristics of TCM such as complexity, holistic, nonlinear, dynamic and so on, so LYFWS research also needs to take the road of complex systems research. Because dialectical and diagnosis standard of LYFWS is more, and appears a high dimensional state after digitized, we want to reveal its dialectical regulations using ideas and methods of modern theoretical modeling from the angle of scientific theory.Based on the above guiding ideology, this paper analyzed LYFWS to establish the best prediction model by using MARS algorithm. The main contents are as follows:(1) Analyzed the advantages of MARS in high dimensional modeling and it’s different from traditional methods. It has a good ability to explain model which indicate the classification of variables. Besides, the model can be described more intuitively under the circumstances of ANOVA decomposition.(2)According to the complexity of TCM system, we mined relationship between variables, looked for evidence of implicit type model after symptoms data quantified and standardized. And then analyzed the internal structure of relationship between the symptoms and syndromes, symptoms and symptoms obtained from TCM interrogation to establish model;(3)We intuitively observed predictor variables using ANOVA decomposition figure of model, analyzed the way that variables affect model to predict the way that symptoms affect LYFWS;(4)We tested model with the reserved data and evaluated precision of the model by specified indicators. The finally results show that the model of LYFWS has good prediction accuracy.