Feature selection: methods depending on data types

Casper72

Dear fellow Rapidminers,

I am trying to predict a binary dependent outcome of a large (80.000 obs.) dataset with 210 possible predictors. Before attempting any backward elimination or maybe even brute force methods I would like to identify the most useful variables to reduce computational time. The variables are both continuous measurements, categorical ordinals and nominals. My knowledge is limited, but I think it would probably be appropriate to use different methods of feature selection depending on the data type?

What kind of operators would you use for each of the data types (Weight by information gain, Chi-Squared etc.)? Or should I contemplate a completely different approach of feature selection?