Machine Learning:

Random Forest Classification

4 Key Pieces

1. Data split

​

• Train models

​

• Models make Predictions

​

• Majority Wins

2

1. Split the Data

Sampling with Replacement

Split the Data

• The full dataset gets split into a number of smaller datasets

​

• Each of these smaller datasets have small differences, making them more diverse

​

• Sampling is done WITH replacement so some data points may be repeated/left out

4

2. Train Models

Train models on their respective datasets

Train Models

• Each sample will have a decision tree trained on it

​

• Trees are trained in parallel (independent of one another)

​

• Vocab:
• Branch (the lines indicating splits)
• Leaf node is a node with no branches

​

​

6

3. Models Make Predictions

Each model makes its own prediction from its training

Models Make Predictions

• Each model based on its own training on its own subset of the dataset makes a prediction

​

• These predictions are a classification

​

• Different trees may make different class label predictions

​

8

4. Final Prediction

Majority Wins

Majority Wins

• The prediction of each model is factored in

​

• The class label that is most predicted is the final output of the model

​

• The model output is what the model predicts is the classification (not guaranteed to be the actual class label)

10

Practice Tree

11

12

13

Three different trees – take majority vote

Now…

Let’s Dance!!!

14