Can model complex and non-linear data

The results are difficult to understand

Numerical type

Nominal type

Data collection

Prepare data

analyze data

training algorithm

Test algorithm

Use algorithms

Split too quickly

Cannot handle continuous features directly

Divide the data set into two parts each time

Easy to adjust during tree construction to handle continuous features

Saves build time

Features to be segmented

Feature value to be segmented

right subtree

left subtree

If the node cannot be divided, it is stored as a leaf node.

Perform binary splitting

The right subtree calls createTree()

The left subtree calls createTree()

square/absolute value

Split the data set in the best way

Generate corresponding leaf nodes

Generate leaf nodes

mean of target variable

var()

for each feature

Returns the features and thresholds for the best segmentation

All values ​​are equal

The effect is not much improved after segmentation

The size of the two subsets after segmentation, one is smaller than the user-defined minimum number of samples for segmentation

The algorithm is very sensitive to the input parameters tolN and tolS

The data set needs to be divided into test set and training set

First, specify the parameters so that the constructed tree is large enough and complex enough to facilitate pruning.

Then, find the leaf nodes from top to bottom, and use the test set to determine whether the merger can reduce the test error. If so, merge

Split data based on existing trees

Returns the tree average

A large number of nodes were pruned, but not into two parts

For a given data set, first fit a linear model

Then calculate the difference between the true target value and the predicted value

Summing the squares of these differences gives the desired error

Create a raster image from an image