Measure Impurity in a Split
A tree splits past deliveries into route groups.
Measure Impurity in a Split explained visually
A tree splits past deliveries into route groups. English captions are optional and start off. The complete transcript follows the written lesson.
What you will learn
A tree splits past deliveries into route groups.
A practical question
A decision tree considers splitting delivery records by route. How do we tell whether a proposed split helps? Start by looking at how mixed the target labels are before and after.
The useful idea
A group containing only on-time trips is pure for that label. A group evenly divided between on-time and late trips is much more mixed. An impurity measure turns those proportions into a number. It doesn't judge whether route is fair or causal; it summarizes the training labels in that group.
Work through the example
Compare a node with ten on-time trips and none late to one with five of each. Then test what happens if a proposed split sends most records to one child and just one example to another. Child sizes matter when we score the overall improvement.
Limits and next step
Next, we'll combine those child impurities into an information-gain calculation.
Worked example
Compare a node with all on-time deliveries to one split evenly between late and on-time.
Quick check: test your understanding
Try it yourself: In the situation described above, explain the method and one reason its output might be misleading.
Check the reasoning
Impurity measures how mixed target labels are within a candidate group. A pure training node with very few records may still generalize poorly.
What is the main idea behind measure impurity in a split?
Impurity measures how mixed target labels are within a candidate group.
How is it used in the example?
Compare a node with all on-time deliveries to one split evenly between late and on-time.
What limitation should be checked?
A pure training node with very few records may still generalize poorly.
Further reading
Machine Learning — Tom M. Mitchell.
Original ToolsInVision teaching examples. Editorial approach · Suggest a correction
Accessible referenceLesson 25 transcript136 words · Video transcript · Matches the published narration and captionsOpen transcript
A decision tree considers splitting delivery records by route. How do we tell whether a proposed split helps? Start by looking at how mixed the target labels are before and after.
A group containing only on-time trips is pure for that label. A group evenly divided between on-time and late trips is much more mixed. An impurity measure turns those proportions into a number. It doesn't judge whether route is fair or causal; it summarizes the training labels in that group.
Compare a node with ten on-time trips and none late to one with five of each. Then test what happens if a proposed split sends most records to one child and just one example to another. Child sizes matter when we score the overall improvement.
Next, we'll combine those child impurities into an information-gain calculation.