An example of Data Mining

Simplifying Heuristic Rules

Fishing in Brasil

Suppose you have a table of rules like this classic example from J.R. Quinlan:

Line	Weather	Temperature	Humidity	Windy	Fishing

1	sunny	hot	high	no	bad
2	sunny	hot	high	yes
3	rain	cool	normal	yes
4	sunny	mild	high	no
5	rain	mild	high	yes
6	cloudy	hot	high	no	good
7	rain	mild	high	no
8	rain	cool	normal	no
9	cloudy	cool	normal	yes
10	sunny	cool	normal	no
11	rain	mild	normal	no
12	sunny	mild	normal	yes
13	cloudy	mild	high	yes
14	cloudy	hot	normal	no

Pacu

The rules above can be rewritten in reduced normal form like this:

Weather	Temperature	Humidity	Windy	Fishing

sunny		high		bad
sunny	hot
	hot		yes
rain			yes
sunny	mild		no
cloudy				good
rain			no
sunny	cool
sunny		normal
sunny	mild		yes
	mild	normal
	hot	normal
		normal	no
	cool		no

Apapa

Each reduced rule represents one or more original rules, denoted by the line numbers of the first table. The set of reduced rules may be parially ordered, as shown here:

Implies	Implies	Implies	Fishing
2	1,2	1,2,4	bad
	4	1,2,4
		3,5
	14	6,9,13,14	good
		7,8,11
	10	10,12
	12	10,12
	12	11,12
	14	8,10,11,14
10	8,10	8,10,11,14

Jacunda

Because of the ordering, the most general rules above are sufficient. Furthermore, a subset of those covers all rules in the original table:

Antecedent	Fishing
1, 2 ,4	bad
3, 5	bad
6 ,9 ,13 ,14	good
7, 8, 11
10, 12

Bagre

Replacing the line numbers of the original table with the reduced rules:

Weather	Temperature	Humidity	Wind	Fishing

sunny	high		bad
rain		yes	bad
cloudy			good
rain		no
sunny	normal

Recolhete