Robotics Reference
In-Depth Information
teaching a program a set of such rules. Niblett and Shapiro tested ID3
on the endgame of king and pawn versus a lone king, and found that
the decision trees generated by the algorithm were 100 percent accurate
when validated against a complete database of such positions.
This early work of Quinlan's, and an improved algorithm which he
called C4.5
46
and published in 1992, served as the foundation for much
of the subsequent research into data mining, a discipline that deals with
extremely large amounts of data and which usually employs some form of
automated learning. For many years the commercial organisations that
offered data mining services tended to make exaggerated claims. One
apocryphal tale describes how data mining, applied to the contents of
each of its customers' shopping baskets, alerted a major supermarket
chain to the fact that sales of baby diapers and beer were highly corre-
lated. This was assumed to be because young fathers who dropped in at
its stores on their way home from work to pick up supplies of diapers,
often decided to stock up on beer at the same time. The supermar-
ket chain then, supposedly, put the two items side-by-side on its shelves,
whereupon sales soared—an excellent advertisement for data mining, not
allowing the truth to stand in the way of a good story.
But in recent years the claims from the data mining industry have
increasingly been true. Thanks to bigger and faster computer hardware
and to advances in data mining algorithms, fiction has turned into fact.
A British supermarket that had just about decided to discontinue a line of
expensive French cheeses that were not selling well, discovered through
data mining that the few people who did buy those cheeses were among
the supermarket's most profitable customers, so the cheeses acquired a
fresh lease of life in order to retain the custom of these big spenders.
Expert Systems
Expert systems
47
are sophisticated computer programs that use human
knowledge to solve problems normally requiring specialist expertise.
46
C4.5 starts with large sets of cases belonging to known classes. The cases are scrutinized for
patterns that allow the classes to be reliably discriminated. Those patterns are then expressed as
models, in the form of decision trees or sets of if-then rules, that can be used to classify new cases,
with the emphasis being on making the models understandable as well as accurate. The system
has been applied successfully to tasks involving tens of thousands of cases described by hundreds of
properties.
47
Also known as “Knowledge-based Expert Systems” and “Intelligent Knowledge-based Systems”.
