For example why you choose IDW, kiriging ... etc. and how many samples that should be considered for evaluating the method or comparing between two methods in interpolating?
You're talking about interpolation. Fundamentally, you need to be sure that the interpolation method you are applying matches the nature of the data you are applying it to. How does the data change with distance? Does the interpolator capture that change? Kriging is flexible in this regard because you tailor its function to match the data. To use it effectively, you need to be able to understand the tool's function - research is the only way to do it in the first instance.
In order to get more accurate results, I suggest that you divide your surface layer into sub-layers based on homogeneity, then apply the IDW for each sub-layer separately, then combine them again.
Also, the Geostatistical Analysis tool allows you to create a random subset of your data points. (validation data) The rationale of this is to interpolate using the interpolator of choice and use the subset to validate the surface, which quantitatively measures the surface compared with the data that was left out of the interpolation. This can be a very powerful tool because it allows you to repeatedly interpolate with different parameters and then decide on the interpolator and parameters that generate the most accurate surface. The final interpolation is then performed on the entire dataset.