I am trying to build a random forest regression model in R using all continuous panel data. There is a large amount of data relative to the number of predictors. Say 500,000 rows and about 40 predictors.
In this case, one of the variables is a t-stat of the correlation coefficient from another regression. However the value given only exists if the t-stat is considered significant (> than a critical value). If the t-stat was not, it comes through as 0.
Looking at the data shows about half of the 500,000 rows are 0, and the rest are populated. I am unsure as to what effect a variable like this would have on a Random Forest Regression, with so many 0 values. Are there usually issues with accuracy, or does it simply mess up importance measures? Perhaps in this case the variable should be categorical? (i.e. signficant or not). Some initial google searches hasn't yielded me anything yet.