Data preparation
The data exploration stage helped us identify all the issues that needed to be fixed before proceeding to the modeling stage. Each individual issue requires careful thought and deliberation to choose the best fix. Here are some common issues and the possible fixes. The best fix is dependent on the problem at hand and/or the business context.
Too many levels in a categorical variable
This is one of the most common issues we face. The treatment of this issue is dependent on multiple factors:
If the column is almost always unique, for example, it is a transaction ID or timestamp, then it does not participate in modeling unless you are deriving new features from it. You may safely drop the column without losing any information content. You usually drop it during the data cleansing stage itself.
If it is possible to replace the levels with coarser-grained levels (for example, state or country instead of city) that make sense in the current context, then usually that is the best way...
