Randome Forest

Ensemble Learners

Random Forest is an Ensemble Learner(Bagging, not Boasting). Basically, there are 3 types of ensemble learners, i.e. Bagging, Boosting and Stacking. A short video explanation can be found here:

• Ensemble Learners 2
• Bagging, reduce variance.
• Boosting, reduce bias.
• [Stacking], combine several weak learners(often diffrent in nature) and make it a stronger learner by finding the best way to weight each weak learner.

Pros

• The bias remains same as that of a single decision tree. However, the variance decreases and thus we decrease the chances of overfitting.
• A quick and dirty way out, random forest comes to the rescue. Don't have to worry much about the assumptions of the model or linearity in the dataset.

Cons

• Random forests don't train well on smaller datasets.
• There is a problem of interpretability with random forest.
• The time taken to train random forests may sometimes be too huge.
• In the case of a regression problem, the range of values response variable can take is determined by the values already available in the training dataset.

Refs

Pros and Cons are based on this post.