This paper presents a novel procedure to improve a class of learning system
s known as lazy learning algorithms by optimizing the selection of variable
s and their attendant weights through an artificial neural network and a ge
netic algorithm. The procedure utilizes its previous knowledge base-also ca
lled a case base-to select an effective subset for adaptation. In particula
r, the procedure explores a space of N variables and generates a reduced sp
ace of M dimensions. This is achieved through clustering and compaction. Th
e clustering stage involves the minimization of distances among individuals
within the same class while maximizing the distances among different class
es. The compaction stage involves the elimination of the irrelevant or redu
ndant feature dimensions.
To achieve these two goals concurrently through the evolutionary process, n
ew measures of fitness have been developed. The metrics lead to procedures
which exhibit superior characteristics in terms of both accuracy and effici
ency. The efficiency springs from a reduction in the number of features req
uired for analysis, thereby saving on computational cost as well as data co
llection requirements. The utility of the new techniques is validated again
st a variety of data sets from natural and commercial sources. (C) 2000 Pub
lished by Elsevier Science Ltd. All rights reserved.