ML2 Comparisson of Models

What is a key advantage of Naïve Bayes?

• Processes all cases at root

• Handles real-valued parameters well

• Requires feature transformation

• Robust to irrelevant features

When is Naïve Bayes optimal as a classifier?

• When the problem depends on many features

• When linear combinations of features are critical

• When the assumption of independence holds

• When the model depends on summing contributions of many attributes

What is an advantage of Decision Trees?

• Recursive partitioning is relatively fast

• Requires feature transformation

• Complex to explain

• Handles real-valued parameters well

When is XGBoost recommended to be used?

• When good performance is needed at the cost of computation

• When model explainability is crucial

• When regression works well

• When overfitting is a big concern

What is a key feature of Random Forests?

• Increases Variance

• Dependent classifiers

• Handles overfitting well

• Requires feature transformation

Which ones are true?

• Bagging handles overfitting, Boosting reduces bias

• Bagging is independent classifiers, Boosting is sequential classifiers

• Bagging is sequential, Boosting is parallel

• Bagging reduces bias, Boosting reduces variance

When do we use LDA?

• When Logistic Regression is stable

• When the number of samples is small

• When there are only two classes

• When Logistic Regression doesn't work

What is a strength of SVM?

• Can handle nonlinear feature interactions

• Can handle large feature space

• Can handle linear feature interactions

• Can handle small feature space

What is a disadvantage of KNN?

• Slow at query time

• Easily fooled by irrelevant attributes and the scale of the data

• Optimization and training required

• Handles large feature space well

When is KNN recommended to be used?

• Lots of training data

• Always scale / standardize the features

• Small Training Data

• When there are less than 20 features