Abstarct: In regression analysis we often come across datasets in which some variables are irelevant to the response variable. In order to identify relevant variables and at the same time to eliminate irelevant variables, there are several methods for selecting variables baxsed on linear models or generalized linear models. Although these methods are mostly baxsed on considering a specific model of the set for the data, we must admit that it is not necessary to govern a particular model for the variable selection, or that our prior information fits the model correctly. This is not the case with the free variable model selection.A Model- free variables selection is often performed by variable screening techniques, which often predict the importance of a predictor variable over the response variable. from a statistical dependency perspective.Pearson's correlation coefficient and mutual information are two statistical dependency criteria,A model-free variables selection It is baxsed on the spatial correlation and the interactions that we present in this thesis. The presented algorithm is executed by two real examples, and the results show that the algorithm performs best in finding the variables that are most relevant to the response variable, and at the same time, eliminates redundant variables.