Hello everyone, welcome to the course of machine learning with Python. In this video, we shall learn about random forest classifier. So when you shouldn't pre fills an ideal classifier should be able to minimize both error in training and test data sets. But decision trees are prone to overfitting, especially when a tree is particularly deep in pursuit of designing a perfect tree, which classifies the training data very accurately, but fails to generalize on unseen data or test data overfitting results in a decision tree which is more complex than necessary. One way to deal with overfitting problem of decision tree is to design an ensemble of decision trees popularly known as random forest. So what is random forest?

It is an ensemble classifier made using many decision tree models to make better and more accurate predictions. It emphasizes on creating a strong and more accurate classifier by combining many weak learners or weak classifiers. So, random forest. So, random is basically due to the trees that is the training set subsets and the features are selected at random with replacement from the original data set and forest because it is a collection of decision trees, now, inclusion of random forest is something like this. So, we have the large data set and we are selecting the subset of the data sets. Now, each of these subsets are being fit to several decision trees as we can see over here.

So, this offset has been feed over this decision tree the subsidies feed over this decision tree and there are so many decision trees now when this sample comes based on the rules of each of the decision trees each of the decision tree will produce some output. Now, we will take majority voting out of all the outputs of the decision trees inside this random forest. And we will provide the final class level based on the majority 40. So this is the intuition of random forest. So how do we create random forest? First is bootstrapping.

So bootstrapping is a kind of resampling, where large numbers of smaller samples of the same size are repeatedly drawn. With replacement from a single original large data set pre bagging randomly select key features from total p features in the data set. Easily K is very less than P. The feed decision trees we feed the decision trees for each of the bootstrap sample and with reduced number of randomly selected features. After creation, we will predict for a test sample so take the test sample And use the rules of each randomly created decision tree to predict the outcome and store the target. Calculate the vote for each predicted target. Consider the majority voted predicted target as the final prediction from the random forest algorithm.

Know how many number of decisions trees should be there in a random forest. The number of decision trees to decide for a random forest is problem specific. It depends largely on number of data records in a training data set. And the number of feature as the output of that estimate is computed to decide on the number of decision trees which will yield the optimum results out of acceptors are those which are not sampled and not fitted by any decision tree inside the random forest. Now for a particular problem, we create in random forest, let's say n equals to 10. And with different number of queries, let's say Number of priests total number of priests 30 number please 50 number please hundred number of priests and so goes on we record they're out of bag error rate and see the number of priests were out of that error rate stabilizes and reaches minimum.

Finally, we choose that random forest for our particular problem, we have to keep in mind that the number of fish should be small as possible. And at the same times out of bag error should be also smaller and minimum advantages. It provides better accuracy as compared to a single decision tree, as it inherently reduces overfitting it runs efficiently on large data set random forests produces very competitive result as compared to many state of the art classifiers these advantages, so there is only one disadvantage of random forest that it takes longer time to train and test that single decision tree. So so far, this is our discussion on random forest. In the next video, we shall discuss on how to implement random forests in Python. So see you in the next lecture.

Thank you.