For complex database prediction problems such as medical diagnosis, th
e ARTMAP-IC neural network adds distributed prediction and category in
stance counting to the basic fuzzy ARTMAP system. For the ARTMAP match
tracking algorithm, which controls search following a predictive erro
r, a new version facilitates prediction with sparse or inconsistent da
ta. Compared to the original match tracking algorithm (MT +), the new
algorithm (MT -) better approximates the real-time network differentia
l equations and further compresses memory without loss of performance.
Simulations examine predictive accuracy on four medical databases: Pi
ma Indian diabetes, breast cancer, heart disease, and gall bladder rem
oval. ARTMAP-IC results are equal to or better than those of logistic
regression, K nearest neighbour (KNN), the ADAP preceptron, multisurfa
ce pattern separation, CLASSIT, instance-based (IBL), and C4. ARTMAP d
ynamics are fast, stable, and scalable. A voting strategy improves pre
diction by training the system several times on different orderings of
an input set. Voting, instance counting, and distributed representati
ons combine to form confidence estimates for competing predictions. (C
) 1998 Elsevier Science Ltd. All rights reserved.