Evolutionary Regression was designed to generalize dependencies, where all the input factors and goal parameter belong to a continuous interval. We are going to use models of such dependencies as invariants in the knowledge system.

Today we poses an algorithm and its realization in the form of Java/C++ class libraries, where generalization of a continuous dependency is performed as a single function for the whole area of input variables existence Y = f(x). The algorithm is looking for an optimal function f(x) in the class of a polynomials of a given power. In case of a high number of dimensions Evolutionary Search is used to minimize structure of the polynomial.

You can explore abilities of the Evolutionary Regression or even build a regression model from your own data at: 

Our testing proved convergence of the algorithm to the polynomial of optimal structure and approximation error. In other words, if a dependency in data can be represented by a polynomial of input variables then Evolutionary Regression algorithm will find the polynomial.

It's quite another matter if a dependency exists but cannot be approximated by a sole polynomial common for the whole domain of existence. We shall have to develop more flexible algorithm for multivariable dependencies approximation or find a way to split domain of existence and build individual models for every sub-domain.