1

I am creating an ensemble classifier consisting of 5 different classifiers. Each of these individual classifiers has an accuracy of 80%. What then, will be the accuracy of the ensemble classifier?

My professor said that it would be around 94%. And, he said that we are free to assume that each of the individual classifiers is independent.

Going by this SE post, I multiplied all the individual accuracies and got a final accuracy of 32.78% - different than what the professor got.

Could someone please explain how the professor got his accuracy? (Cannot ask him since he's currently unavailable). Also, as you might have guessed, I am new to probability. Hence, requesting the help.

Thanks!

Someone
  • 143

1 Answers1

1

Since the classifiers are independent and there is no information to suggest some are preferred over others, the ensemble classifier is the majority decision of the 5 classifiers. The ensemble accuracy is therefore the probability that 3 or more of the 5 classifiers make the correct classification. The number of classifiers that make the correct classification is given by the binomial distribution with 5 trials and success probability 0.8. Sum that distribution from 3 to 5, and you will get 0.94208.

Dean
  • 1,370