Math question - how to remove values way different from others in an array of numbers

Lets say in an array I have 20 items

12834
34252
53526
12320
12140
353622
534
53277
89233
23512

how can i 'detect' that 353622 is wildly different from others in the list, as is '534'? And if i wanted to be mroe stringent.. detect '89233' as different too?

What are the rules that define something as different?

Quote:

---

Originally Posted by MartinLiss

What are the rules that define something as different?

---

Im not sure.. these would be defined I suppose as 'outliners' via statistics

Maybe someone here can help.

Let me see how much I can remember from statistics.

You would want to find the mean (sum of numbers divided by the number of numbers (n)) however if you have significant rouge numbers you may wish to multiply all the numbers and find the nth root.

In your case the tenth root.

Now you have a reliable mean you can work out the standard deviation. The formula for that should be easy to find in Google (I can't remember it off the top of my head). Spread sheets packages have such functions built in...

Then you would want to find every thing within one standard deviation which should be around 50% of your population (those numbers).

This page seems to cover that pretty well:

http://www.statcan.ca/english/edu/po...2/variance.htm

Now you would be looking for a number of standard deviations that contain 95% or more of your population sample. As your sample is so small you are going to be looking for contains all but 2 or 3.

inside_range > (n - 3)

Using code you will have an upper bound and a lower bound outside of which are values you don't want. You could drop these from the array and recalculate the mean and the SD and repeat until you are happy.

The other option would be create a function that takes the mean and SD and a value as arguments and returns the number of SDs the number is from the mean. Again you would have to look up such a formula but it is fairly simple stuff I recall. You could then test for values that have a SD that is outside of the acceptable range and drop them at that stage.

I hope that helps.

yeah i was thinking something similar.

First, take the mean of all the numbers
Second, find the range of all the numbers
Third, create a new range based on a preset % of the old range
new_range = [mean-%*old_range/2 mean+%*old_range/2]