Hi Swathy

What you are seeing is common (and maybe even expected). The link below shows that using different tests for optimal number of clusters gives different numbers - and this using the same data and the same k-means method. I would take the view that your results (2 and 3) are suggestions which can't be internally validated. You would normally choose using external factors (meaning: what problem are you trying to solve? How are you reporting results? etc.) and note in your writeup that both 2 and 3 were valid choices based on the data alone.

The link below is very high quality (in my opinion), and gives much more useful detail to what I've just written:

https://towardsdatascience.com/10-tips-for-choosing-the-optimal-number-of-clusters-277e93d72d92

Thank you, Professor Thomas W Kelsey ! I found it helpful.