Build the training dataset correctly
The list containing the training lines data was constantly re-initialized after each page. That explains the really low training sample (and fast training time as well I guess).
The list containing the training lines data was constantly re-initialized after each page. That explains the really low training sample (and fast training time as well I guess).