مقارنة بعض طرائق االنحدار الالمعلمي الجمعي

Abstract

In the absence of knowledge about the phenomenon was the experience for the first time or can not determine a causal relationship, or behavioral variables that link instead of the requirement to take the data template or form the described function phenomenon in advance are replaced with a more flexible manner so-called Nonparametric analysis.The expansion in the Spline Smoothing Situation from Single to multiple variables showed the problem of dimensionality because we must expand in the case of Cubic Spline Smoothing to the state of Thin Plate Spline and the method of analysis of this type of Splines difficult, especially if he had known extent of interaction and which can not be represented easily as needs a high level of analysis and programming and this was the idea of Additive Model models and especially that there are some algorithms to overcome the effect of dimensionality problem.Research aims to focus on the methods Nonparametric Additive Regression Methods (i.e., the case of binary variables) so that we avoid the problem of dimensionality and the algorithms used Backbiting and SIMEX, which combines simulation and interpolation, and comparison between the algorithms using the standards of existing and other proposed criterion MGCV, using many of the simulation experiments and variations and sizes of samples different, in addition to the application of algorithms on real data about water pollution and the adoption of the model that best fits the data, which gives the less comparison criteria.