Learning Decision Trees

• When I think of learning decision trees, I think of C4.5. This is closely related to what Russell and Norvig describe.
• The key bit is to take the data, and see what feature splits apart the data best.
• Yellow 2:3, Purple 2:3; Small 2:2; Large 2:4; Stretch 4:2; Dip 0:4; Adult 4:2; Child 0:4.
• In this case, both Age and Act split the data perfectly.
• Train

  Colour	Size	Act	Age	Category/Inflated
  YELLOW	SMALL	STRETCH	ADULT	T
  YELLOW	SMALL	DIP	ADULT	F
  YELLOW	LARGE	STRETCH	ADULT	T
  YELLOW	LARGE	STRETCH	CHILD	F
  YELLOW	LARGE	DIP	CHILD	F
  PURPLE	SMALL	STRETCH	ADULT	T
  PURPLE	SMALL	DIP	ADULT	F
  PURPLE	LARGE	STRETCH	ADULT	T
  PURPLE	LARGE	STRETCH	CHILD	F
  PURPLE	LARGE	DIP	CHILD	F

• I picked Stretch, and if it's False, it gives the correct answer.
• Pull all of those out, and do it again.
ColourSizeActAgeCategory/Inflated
• Modified Train

  Colour	Size	Act	Age	Category/Inflated
  YELLOW	SMALL	STRETCH	ADULT	T
  YELLOW	LARGE	STRETCH	ADULT	T
  YELLOW	LARGE	STRETCH	CHILD	F
  PURPLE	SMALL	STRETCH	ADULT	T
  PURPLE	LARGE	STRETCH	ADULT	T
  PURPLE	LARGE	STRETCH	CHILD	F

• In this case age splits perfectly, and you get the decision tree.
• If a feature is integer or floating point you may need to pick ranges.