Induction of decision rules within the dominance - based rough set approach to the multicriteria and multiattribute classification is considered. Within this framework, we discuss two algorithms: Glance and an extended version of AllRules. The important characteristics of Glance is that it induces the set of all dominance - based rules in an incremental way. On the other hand, AllRules induces in a non-incremental way the set of all robust rules, i.e. based on objects from the set of learning examples. The main aim of this study is to compare both these algorithms. We experimentally evaluate them on several data sets. The results show that Glance and AllRules are complementary algorithms. The first one works very efficiently on data sets described by a low number of condition attributes and a high number of objects. The other one, conversely, works well on data sets characterized by a high number of attributes and a low number of objects. © Springer-Verlag 2004.
Incremental versus non-incremental rule induction for multicriteria classification
Greco, Salvatore
;
2004-01-01
Abstract
Induction of decision rules within the dominance - based rough set approach to the multicriteria and multiattribute classification is considered. Within this framework, we discuss two algorithms: Glance and an extended version of AllRules. The important characteristics of Glance is that it induces the set of all dominance - based rules in an incremental way. On the other hand, AllRules induces in a non-incremental way the set of all robust rules, i.e. based on objects from the set of learning examples. The main aim of this study is to compare both these algorithms. We experimentally evaluate them on several data sets. The results show that Glance and AllRules are complementary algorithms. The first one works very efficiently on data sets described by a low number of condition attributes and a high number of objects. The other one, conversely, works well on data sets characterized by a high number of attributes and a low number of objects. © Springer-Verlag 2004.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.