chengds added The_selection_process_is_clearly__.tex  over 8 years ago

Commit id: 13cda5e655bc1d00d91cfdffb56f242d4ac420ef

deletions | additions      

         

The selection process is clearly imprecise in the case of the self-assessed traits, because two quartiles should always contain half of the samples. With the exception of EXT(S), the selected number of users is much bigger than 150, while it is perfectly balances on the attributed traits. We do not know if this affects the training of the classifiers. Overall, the resulting classes have a balanced number of users, so no classifier should be overtly biased towards one class.