2

I'm designing a machine learning system in which i'm training 4 different classifiers and voting among it's predictions to decide on an output. Presently i'm just making a simple vote (i.e. selecting the most recurring predicted class) but can't I do something better using the outputted probabilities of the models as a sort of weight to decide on the best class?

thiago_lira
  • 41
  • 3

1 Answers1

2

can't I do something better using the outputted probabilities of the models as a sort of weight to decide on the best class?

Yes, see ensemble learning for ideas. E.g. stacking sometimes outperforms the majority vote scheme.

Franck Dernoncourt
  • 42,093
  • 30
  • 155
  • 271