Mohamed, the missed data never should be equal zero.

You can interpolate when your data are afected by spatial/ temporal autocorrelation by using its neighbors. In climatology, you can use missing values based on the median of corresponding values of the same time period.

The situation is different when we are working using experimental data. The regression methods permit to determine missing values. You can remove rows. Only in extreme cases, I could suggest to remove a particular variable.

How to manage missed data?

If you have very large sample, perhaps you can remove the records / rows with missing data.  But when your sample is small, you might want to substitute with mean / median as appropriate.  Some online researchers try to prevent  missing data by configuring their online survey with all questions need to be filled by respondents.

If missing data is too numerous (say more than 25% of all data), the data processing should not be done. Otherwise, some techniques could be used with care (interpolation, substitution by apporpriate statistics, etc.). Note that SAS (Statistical analysis software) has a very good treatment of missing data.

In SAS a dot (.) indicates missing data and SAS correctly takes into account missing data in all computations. In meteorology, missing data are labelled -9999 to clearly identified them.

Some references to help you here:

http://handbook.cochrane.org/chapter_16/16_1_2_general_principles_for_dealing_with_missing_data.htm

http://www.measuringu.com/blog/handle-missing-data.php

https://en.wikipedia.org/wiki/Missing_data

In R, missing values are explicitly represented by the symbol NA.

In another context, into a Geographic Information System, I specify the missing values as -9999 for the raster data. This data represent the data mask to analysis.

There have been major developments in principled ways to handle missing data in a rigorous fashion; this site is an excellent guide

http://missingdata.lshtm.ac.uk/