Skip to content

Ensembling for multiclass classification and regression problems

Notifications You must be signed in to change notification settings

theidentity/Ensembling_Techniques

Repository files navigation

Ensembling_Techniques

Ensembling for multiclass classification problems

Methods Used

  1. Mean Ensembling

    • Takes the mean of the probability if the output classes
    • Output is the class with maximum probability
  2. Majority Voting

    • Assigns as output whichever class has maximum number of votes among ensemble models
    • In case of a tie, takes the first group of tied classes

Results

Model Accuracy
RESNET50 0.969882729211
XCEPTION 0.945895522388
CAPSNET 0.653251599147
CNN_CUSTOM 0.861407249467
MAJORITY VOTING 0.967750533049
MEAN ENSEMBLE 0.982675906183

Ensembling for regression problems

Methods Used

  1. Mean Ensembling

    • Takes the mean of the predicted outputs by each model
  2. Stacking

    • Takes the output of the ensembles and passed it through a MLP Regressor
    • Output is the output provided by the regressor on passing the test data

Results

Model MSE R2_Score
MLP 0.0103210967873 0.808796324494
RBF 0.0106966198534 0.801839564767
MEAN ENSEMBLE 0.00845186510476 0.843424811702
STACKING ENSEMBLE 0.00729725486593 0.864814565717

About

Ensembling for multiclass classification and regression problems

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages