Accuracy Assessment of unsupervised classification using clusterers

[Jyoti]

Hello everyone,

I am trying to classify a district-level median composite sentinel-2 imagery using different clustering techniques available in Google Earth Engine. 

I was successful in classifying the image in 8 classes but I am stuck at the accuracy assessment part for the classification. 

I couldn't find anything related in the documentation too.

It would be really helpful if anyone can help me through this. 

Here is my code: https://code.earthengine.google.com/63a9cc450d5db3d1f3e5ed104d999229

Thanks,

Jyoti

[Pradeep]

Looks like you are doing unsupervised classification (line 88: training samples have no ground truth labels, and you are using clusterers) in which case accuracy assessment is not straightforward if you don't have ground truth reference data.

Do you mean to use the Copernicus landcover in your script as ground truth, or have any of your own ground truth data? If so, perhaps your should consider supervised classification. There are a couple of nice explainers on it, including accuracy assessment, by MD Madhusudan, if you haven't seen them -- https://www.youtube.com/watch?v=KtqKwddKTgk and https://docs.google.com/presentation/d/145Tcs2XX_SfxYGKdjyKY5l9KlP5TplxwEFFtQ_iXjlA/edit