The original simulation was run with actual frequency calculated without Remove NA, which included incomplition data. So after re-run, the results are still not as expected

Consider as 57 classifiers

## [1] "mean value is 0.933248299319728"
## [1] "std is 2.33351595946682"

Consider as 19 classifiers

## [1] "mean value is 0.656462585034014"
## [1] "std is 2.61851869243483"