Algorithm of Building Regression Decision Tree Using Complementary Features

In the so-called explained artificial intelligence, there is a need to build small models, but accurate and intuitive for the analyst. It is necessary to formalize, which models are perceived by analysts and decision-makers as intuitively understandable and plausible. It's shown that the use of accumulated information about additional to each other in some sense, complementary features can improve the accuracy of the small regression decision trees, as well as make them more plausible. The formal definition of the complementarities of the feathers is proposed. Algorithm of building regression decision tree with complementary features is presented. Condition of plausibility of two-levels decision tree is described.