Predicting the price of a stock based on historical data.

Predicting the annual income of a person based on their education and employment history.

Predicting whether a credit card transaction is fraudulent based on some attributes.

Predicting the number of cars that will pass through a particular intersection during rush hour.

Naïve Bayes inherently ignores normalization: The classifier only uses the product of likelihoods and priors, so normalization by P(X) is unnecessary for prediction.

P(X) is constant across all classes for a given X: The marginal probability of the features is the same for every class.  Therefore, it does not change the relative ordering of posterior probabilities and can be ignored for classification.

P(X) is always equal to 1: Since probabilities are normalized automatically, dividing by P(X) has no effect on the outcome.

P(X) is usually very small: Because feature combinations are rare, P(X) is near zero, so including it would not materially affect which class has higher posterior probability.

True

False

It selects support vectors based on distance metrics.

It reduces the dataset to a single dimension to simplify computations.

 It is used only for linear SVMs to compute the margin.

It allows computation of inner products in a high-dimensional space without explicitly mapping points.

Between-cluster entropy

Within-cluster sum of squares 

Maximum cluster distance

 Silhouette score

Always quadratic

Always nonlinear

 Based on cluster means

Linear in feature space (unless transformed)

Information Gain Ratio

Misclassification Error

Entropy

Gini Impurity

It prefers splits with many categories.

It is unbiased toward attributes with many levels.

It is based on reduction in entropy.

It minimizes variance directly.

Cluster centroids

Variance minimization

Linear relationship between features and output

Probability using sigmoid function