# Preprocessing of training dataset for machine learning

I'm developing a log analyzer to predict and find errors in an equipment. Each logged data contains the following format:

  Timestamp | Log source | type of message | message


Each entry log I want to represent by one pixel RGB because considering the 24 bits, is possible to represent the last 3 parameters ( Log source, type of message and message), but I don't have bits enough to represent the timestamp ( this data I will represent by the difference o previous timestamp "delta of time"), the resolution of time is second, and sometimes we have a large time between one log of another. The example above illustrate the situation:

This project has the purpose of analyzing the log to find and predict errors, and this preprocessing is used to simplify the data entry to the machine learning algorithm, this is a good way to represent data to an RNN? Or for this kind of problem exist a better way to make an analyses?