The parameters should capture all the important characteristics of the feature we are trying to detect and not capture any of the ones not important to it. For example, the gradient of the increasing interval is an important characteristic for sign language: the speed of movement upwards indicates rapidity of movement - e.g. the distinction between shutting a window (a slow movement of one forearm downwards against the other forearm) and slamming a window (a much more rapid movement). On the other hand, the variance of the signal may not be important to us, as it may be, for example, the result of noise in our measurement system, rather than some important underlying property of the data.