The mean square error is a form of {{accuracy measure}. It is the arithmetic mean of the squares of the errors. That is, if xi is a sequence of 'true' values (perhaps form tariningdata) and pi is a corresponding sequence of estmates/predictions, then:
mean square error = &Sigma (xi − pi)2 ) / N
Sometimes the sqoare root of this used, called (predictably!) the root mean square error (RMS).