cons
- The assumption that features are conditionally independent given the class is not realistic.
- Generalization performance may worse than more sophisticated learning methods.
Types of Naive Bayes Classifiers
- Bernoulli: use a set of binary occurrence features(word presence/absence)
- Multinomial: use a set of count-based discrete features( word count)
1.2 Gaussian Naive Bayes classifiers
- use continuous or real-valued features
- During training, Classifier estimates the mean and standard deviation of each feature for each class.
How it predicts
- For prediction, the classifier compares the features of the new example data point with the feature statistics for each class and selects the class that best matches the data point.
Assumption
- The Gaussian Naive Bayes Classifier assumes that the data for each class is generated by a simple class-specific Gaussian distribution and pick the class that has the highest probability of generating this data point.
Gaussian Naive Bayes classifiers: high-dimensional data.
Bernoulli & Multinomial: text classification( sparse features vector, features are a large number of distinct words)
decision boundary
- the decision boundary between classes in the two-class Gaussian Naive Bayes Classifier. In general, is a parabolic curve between classes. It will be linear when the variance of these features is the same for both classes.