Скачать или смотреть [MXML-8-01] Random Forest [1/7] - Overview, Bagging, Row-subsampling, Column-subsampling

This video was produced in Korean and translated into English. And the voice is generated by AI TTS. The English translation may contain grammatical errors.

This is part 1 of a series on Random Forest, which is the 8th module of machine learning.

Random Forest is an ensemble model that uses multiple Decision Trees. Random Forest has a variety of features, we'll take look at them one by one.

Let's take a look at the full table of contents of Random Forest.

Chapter 1 provides an overview of Random Forest.

Chapter 2 discusses the Bagging based Ensemble learning called Bootstrap Aggregation.

And in Chapter 3, we will look at how to sample data. Random Forest uses column subsampling as well as traditional row subsampling to prevent overfitting.

And in Chapter 4, we will write code that implements Random Forest. Let's implement a Random Forest from scratch using subsampling by rows and columns. We will then implement this using scikit-learn's RandomForestClassifier and compare the results.

In Chapter 5, we will look at Out-Of-Bag score to measure the performance of Random Forest. Let's learn more about OOB through examples and implement it with code.

And in Chapter 6, we'll look at how to handle missing values. Learn about the Proximity Matrix and how to use it to estimate missing values in training and test data. We'll then write code to create a Proximity Matrix and estimate missing values.

In Chapters 7 and 8, we will look at how to detect outlier data points. In Chapter 7, we will learn how to detect outliers using the Proximity Matrix of Random Forest, and in Chapter 8, we will learn how to detect outliers using Isolation Forest. We will then interpret each result and compare the two results.

In this video, we will look at an overview, Bagging, and the data sampling methods,

According to Wikipedia, the Random Forest was first proposed by Tin Kam Ho in 1995. And Leo Breiman, who proposed CART, extended this algorithm in 2001.

Random Forest is a type of Bagging based ensemble model, but uses Decision Trees as the base learner. The variance of the estimation results is small and the possibility of overfitting is low, showing good overall performance.

In Random Forest, data is sampled not only row-wise but also column-wise. Subsampling by rows reduces correlation between trees, and subsampling by columns reduces correlations even further. Column subsampling is the random selection of features.

Make each tree deep enough and do no pruning. Random forest is less prone to overfitting despite using multiple deep Decision Trees. This is because it uses sample data from both rows and columns, and averages the outputs from multiple trees. This eliminates the need to prune each tree.

#RandomForest #RowSubsampling #ColumnSubsampling #Bagging