Batch learning has led to the development of many classifiers in different paradigms, such as divide and conquer, lazy learners, kernel methods, graphics models, and so on. Now, if we move to a stream for the same, we need to understand how to make them incremental and fast for the large datasets in the streams. We have to think in terms of the complexity of the model versus the speed of the model update, and this is the main trade-off that needs to be taken care of.

The majority class algorithm is one of the simplest classifiers, and it is used as a baseline. It is also used as a default classifier for decision tree leaves. Another is the no-change classifier, which predicts the labels for new instances. The Naive Bayes algorithm is known for its low cost in terms of computational power and simplicity. It's an incremental algorithm and is best suited for streams.