Compute Accuracy For Pixel Classification ArcGIS Pro 3.5.1

[Ed_]

Hello ESRI, so this new tool in ArcGIS Pro 3.5 is pretty neat, however I have a couple of questions. I have noticed that the tool returns different accuracy results each time the tool is ran even though the input data (pixel classification raster and ground truth feature class) along with the number of random points stay the same. For example in two instances I got accuracies of 47% and 52%. 

Is this because of the random distribution of points? like the spatial distribution of the random points is different each time the tool is ran?

Which of the above two accuracies are correct then 47% or 52%?

[DanPatterson]

Compute Accuracy For Pixel Classification (Image Analyst)—ArcGIS Pro | Documentation

neither is correct per se, except for the instance that it applies to.

experiment with the sampling strategies as well as the number of points

[PavanYadav]

@Ed_ Yes, it appears to be because of the default 5000 random distribution of points. I feel we should increase the default from 5000 to a larger number.

[PavanYadav · Accepted solution]

@Ed_ 

just wanted to add, there are 3 different modes to create the AA points. please consider the equalized stratified random mode which puts the same number of AA points per class. Then the overall accuracy has each class contribute equally.