| non-gravity said:
( (predicted - real)/real*100% plus (predicted - real)/predicted*100% )/2 and then taking the absolute number in whole points .
example: Predicted 200 Real sales number 250 ( (200-250)/250*100 plus (200-250)/200*100 )/2 = -22,5 = 23 points
|
Can you explain the math for me?
It's the difference between predicted and real, over the real*100 plus the difference over the predicted*100, all divided by 2?
Which means, the percentage your number needed to be the actual plus the percentage the actual needed to be your prediction, divided by 2 for two variables.
Is this a standard measure for determining accuracy?
I understand the second part is standard forecast accuracy equation. What is the first part? ie: why divide by the actual?
Also, because percentages of actual and real scale based on the values of the inputs, isn't it skewed towards things that have higher sales?
predicted: 200, real:250
( (200-250)/250*100 plus (200-250)/200*100 )/2 = -22,5 = 23 points
predicted : 400, real: 450
( (400-450/450*100 plus (400-450/400*100)/2=11.11 12.5 = 11.8 = 12 points
EDIT: ah nevermind the questions about the formula, it's an average to smooth out bias towards under and over predicting.
