Abstract
Two methods from the field of artificial intelligence
were implemented and employed on a medical data set, in
order to perform data mining. The data set consisted of
cases from patients who suffered recurring miscarriage,
and the aim was to investigate whether the implemented
methods were able to identify previously unknown
factors associated with recurrent miscarriage. The
first approach used a specific type of artificial
neural network - Kohonen's self-organizing map for
performing clustering within data sets. By using new
cluster detection methods and the visualisation
possibilities of the employed programming language
Java, and its graphical user interface components
Swing, it allows interactively the visualisation of
relationships within a data set. The second, relatively
unique approach, infers rules from a data set by using
the paradigm of genetic programming. The rules consist
of an IF-part (antecedent) and a THEN-part
(consequent). The system has to be supplied with the
consequent and works out antecedents, which describe
the sub data set indicated by the consequent within the
supplied data set. The antecedents produced take the
form of a tree where Boolean operations AND, OR and NOT
represent nodes, and Boolean expressions represent the
leaves. Boolean expressions can be built from all types
of data including free-text and real numbers. This
system was also implemented with Java and offers in
addition the possibility of knowledge extraction from
clusters built by the self-organizing map approach.
Users
Please
log in to take part in the discussion (add own reviews or comments).