Errors in unsupervised classification (k-means) while using Landsat 5 & 7

[Adil-Aman]

The training sample point dataset is not homogenously distributed in the defined region when using Landsat 5 or Landsat 7. The clustering result is generated on and around the scattered clusters of training points only, not for the whole image. Does anyone else experienced the same issue?
Training dataset:

Clustering results:

[giswqs]

Use stratified sampling or increase the number of sample points.

[Adil-Aman]

Thanks Dr Wu, stratified sampling works, but increasing number of sample points only make those dispersed clusters bigger.
I want to do unsupervised classification. but I think I can't do it using stratified samples.
Any other suggestion to perform unsupervised classification

I experienced same issue with any feature / featureCollection (as ROI), for various regions, various dates, and with or without cloud mask for Landsat 5 & 7. The example above has Landsat 7 image in the background and there is negligible cloud cover. I think it has something to do with pixel values, as most samples are clustered in water, dark pixels, or bright pixels (high vegetation)

[Adil-Aman]

Issue solved. The problem occurred only when using SR data, TOA data for Landsat5 & 7 is working fine