Multiple and complex network built by the path coefficients for partial least squares variable selection research

摘要

The complexity of data processing problem of identification of Chinese herbal medicine, Chinese medicine prescription dose-effect relationship for experimental and literature data, the independent variables and the dependent variable nodes, the direct and indirect path coefficient weights to build a complete bore figure. Complex network model to the right to figure analysis, statistics each variable to point to the path due to the variable point K steps, draw a path different weight, based on the weight value divided on the dependent variable direct and indirect impact the argument points group proposed new PLS-based variable delete a criterion, explore innovative ideas. A PLS variable selection, in order to simplify the discriminate model, including a new algorithm based on error.