On the efficacy of old features for the detection of new bots