Calculating most accurate mean prediction from two ranges of predictions.
Hi all.
I have a program written in visual basic 6 that calculates predictions using a standard statistics equation.. ie YPredicted = Ymean +/- R(Sy/Sx)(X-Xmean)
... the calculations are not the problem
I have a variety of inputs that are used to create the prediction equation, and I want to estimate the accuracy or otherwise of using one or more sets of inputs, so as to see what is the best set of inputs to use in order to obtain the most accurate prediction.
I have another program, that I am designing to see which is the most accurate set of inputs to use, that makes statistically random selections from a database of over 500K rows, selecting, for example 1000 rows.
the program then calculates a number of predictions, each using a variety of inputs, so at the end I have, for each row, an actual figure, and several different predicted values for these same figures.
From each of these values, I can therefore quite easily calculate the error in the prediction for each given method or selection of inputs.
So, I then have a array of actual predictions, and the errors of these predictions, for each method, or group of inputs, for the prediction.
What I want to be able to do is summarise this info.
I guess what I'm asking is:
how can I tell which prediction method is the most accurate for the given sample...
i.e. is it the one where the means of the predicted error is the lowest, or is it the one where the standard deviation of such predictions is the smallest, and includes the actual prediction..
e.g. actual = 100KP/H
Mean Predicted 1 = 101 KP/H, std dev of 5 KP/H
or
Mean Predicted 2 = 102 KP/H, std dev of 3 KP/H
How do I calculate which is the most accurate...
While the first is actually the closest predicted mean to the actual mean, the std deviation is a lot higher than the second.
I have a fairly good knowledge of statistics, but, as I am entirely self taught, I have LOTS of holes in my knowledge.
Any help/advice greatly appreciated...
Thank you
|