It is a well-known problem of milling machines, that waste heat from motors, friction effects on guides and most importantly the milling process itself greatly affect positioning accuracy and thus production quality. An economic and energy-efficient method of correcting this thermo-elastic positioning error is to gather sensor data from the machine tool and the process and to use that information to predict and correct the resulting tool center point displacement using high dimensional characteristic diagrams. On the one hand, the selection of which and how many input variables to use in the characteristic diagrams is critical to their performance. On the other hand, however, there are often a great number of possible variable combinations available and testing them all is practically impossible. This paper will discuss the suitability of many different input variable types and present a new method of input variable selection which will be compared to existing methods and demonstrated on measurements performed on a machine tool.