Gloria Gheno, Transformed Tree-Structured Regression Method, BMSA Volume 16, Bulletin of Mathematical Sciences and Applications (Volume 16)
    Many times the response variable is linked linearly to the function of the regressors and to the error term through its function <i>g</i>(Y). For this reason the traditional tree-structured regression methods do not understand the real relationship between the regressors and the dependent variable. I derive a modified version of the most popular tree-structured regression methods to consider this situation of nonlinearity. My simulation results show that my method with regression tree is better than the tree-based regression methods proposed in literature because it understands the true relationship between the regressors and the dependent variable also when it is not possible to divide exactly the error part from the regressors part.
    Birckel-Docksum Transformation, Nonlinearity, Random Forest, Regression Tree