There are several methods of dealing with missing data fields. The arithmetic mean of each field could be calculated from those records containing a meaningful value, and put to the missing fields. However, this approach might reduce the significance of that particular data field in network training and data vector classification. Or the resulting data records could indicate behavior that does not exist in real life.
One good approach would be to discard just the missing field from each phase of the analysis, if possible. It depends on the analysis method, if this approach can be used or not. At least some neural network methods would allow the exclusion of some fields from all the phases of the analysis.