currently in DES literature, techniques, such as bagging and random subspace are used. however. when such techniques are used, the base classifiers are generated based on a random sampling for training data and there is no guarantee to have diversity between the generated classifiers. can any one propose other classifier generation?
Vikas Ramachandra
random forests are not just randomizing over training data, but also over features, thus reducing the problem of all trees relying heavily on one or few features for splits. this will result in some diversity. you can also built heterogeneous ensembles (each classifier being of a different type such as SVM, RF, etc)
1 votes
0 thanks
- Badges
- Science topic
- Similar topics
- Filing
More Maryam Kaviyani's questions See All
Similar questions and discussions