M.R. Islam and S.M. Khan (Bangladesh)
Data mining, classification methods, distinguishability principle, rule generator, decision tree, feature selection.
In this paper, a distinguishablity principle which can be applied to different classification algorithms is presented. It is designed to find out the best (from the point of view of separation of different classes) divisions of continuous features or the best combinations of symbolic values of discrete features. It has been tested on several datasets to construct classification decision trees and decision rules. It can also be useful in continuous features discretization and estimation of feature's importance, which forms a first step of neural algorithms for extraction of logical rules from data.
Important Links:
Go Back
