Name
Abstract | ||
|---|---|---|
We define a class of patterns generalizing the jumping emerging patterns which have been used successfully for classification problems but which are often absent in complex or sparse databases and which are often very specific. In supervised learning, the objects in a database are classified a priori into one class called positive - a target class - and the remaining classes, called negative. Each pattern, or set of attributes, has support in the positive class and in the negative class, and the ratio of these is the emergence of that pattern; the stimulating patterns are those patterns a, such that for many closed patterns b, adding the attributes of a to b reduces the support in the negative class much more than in the positive class. We present methods for comparing and attributing stimulation of closed patterns. We discuss the complexity of enumerating stimulating patterns. We discuss in particular the discovery of highly stimulating patterns and the discovery of patterns which capture contrasts. We extract these two types of stimulating patterns from UCI machine learning databases. |
Year | Venue | Keywords |
|---|---|---|
2010 | ICCS | negative class,closed pattern,positive class,stimulating pattern,sparse databases,calculation method,target class,classification problem,uci machine,supervised learning,remaining class,machine learning |
Field | DocType | Volume |
Pattern recognition,Computer science,Generalization,A priori and a posteriori,Binary decision diagram,Supervised learning,Contrast (statistics),Artificial intelligence,Formal concept analysis | Conference | 6208 |
ISSN | ISBN | Citations |
0302-9743 | 3-642-14196-X | 0 |
PageRank | References | Authors |
0.34 | 14 | 3 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Ryan Bissell-Siders | 1 | 8 | 1.51 |
| Bertrand Cuissart | 2 | 12 | 0.90 |
| Bruno Crémilleux | 3 | 373 | 34.98 |