When developing models there is always a trade-off between model complexity and model fit. In this paper, a measure of learning model complexity based on VC dimension is presented, and some relevant mathematical theory surrounding the derivation and use of this metric is summarized. The measure allows modelers to control the amount of error that is returned from a modeling system and to state upper bounds on the amount of error that the modeling system will return on all future, as yet unseen and uncollected data sets. It is possible for modelers to use the VC theory to determine which type of model more accurately represents a system.
