A study of methods of classifier construction and updating
A. N. Borisov, Yu. V. Kornienko · Automatic Control and Computer Sciences · 2008
This paper investigates the behavior of the inductive algorithm CART2 applied to constructing a classifier using a modern DBMS and solving problems related to classifier incremental updating. As testing objects we used four databases taken from different knowledge domains. The results of the experiments demonstrate good performance for the incrementally obtained classifier—i.e., for the Hypoteroid database (classification of thyroid body diseases), the identification error amounted to 1.95% (for 2870 samples) versus 1.8% for the classifier constructed using the full teaching selection.