ISOETRP Clustering Algorithm and Its Application to Tree Classifier Design
A new clustering algorithm ISOETRP has been developed. Several new objectives have. been intro-duced to make ISOETRP particularly suitable to hierarchical pattern classification. These objectives are: a)minimizing overlap-between pattern class groups, b) maximizing entropy reduction, and c) keeping bal-ance between these groups. The overall objective to be optimized is
GAIN = Entropy Reduction/(Overlap + 1).
Balance is controlled by maximizing the GAIN. An interactive version of ISOETRP has also been developed by means of an overlap table. It has been shown that ISOETRP gives much better results than other existing algorithms in optimizing the above objectives. ISOETRP has played an important role in designing many large -tree classifiers, where the tree performance was improved by optimizing GAIN value.