Please enable JavaScript.

Coggle requires JavaScript to display documents.

BOOSTING - Coggle Diagram

- - - - Why only DT as a weak learner :question:
        
        https://stats.stackexchange.com/questions/124628/why-adaboost-with-decision-trees#
        
        DTs are easy to tweak as a weak learner, just use stumps i.e with depth 1
        
        easy to control the bias - variance tradeoffs
        
        Boosting works well with non linear models and DTs are non linear classifiers
  - - - How?
        
        By using probability distributions
        
        change the prob of data points as such that the prob will be higher for misclassified data points and less for others
        
        This concept can be embedded into objective function which our algorithm try to minimise, this way our algorithm will put more focus on data which are misclassified