Abstract | ||
---|---|---|
We derive upper and lower limits on the majority vote accuracy with respect to individual accuracy p, the number of classiers in the pool (L), and the pairwise dependence between classiers, measured by Yule's Q statistic. Independence between individual classi- ers is typically viewed as an asset in classier fusion. We show that the majority vote with dependent classiers can potentially oer a dramatic improvement both over independent clas- siers and over the individual accuracy p. A functional relationship between the limits and the pairwise dependence Q is derived. Two patterns of the joint distribution for classier outputs (correct/incorrect) are identied to derive the limits: the pattern of success and the pattern of failure. The results support the intuition that negative pairwise dependence is benecial although not straightforwardly related to the accuracy. The pattern of success showed that for the highest improvement over p, all pairs of classiers in the pool should have the same negative dependence. |
Year | DOI | Venue |
---|---|---|
2003 | 10.1007/s10044-002-0173-7 | Pattern Anal. Appl. |
Keywords | Field | DocType |
Key words: Classifier combination,Classifier fusion,Diversity,Independence and dependence,Limits on majority,Majority vote | Pairwise comparison,Classifier fusion,Joint probability distribution,Voting,Pattern recognition,Upper and lower bounds,Intuition,Q-statistic,Artificial intelligence,Majority rule,Statistics,Mathematics | Journal |
Volume | Issue | ISSN |
6 | 1 | 1433-7541 |
Citations | PageRank | References |
161 | 7.06 | 22 |
Authors | ||
4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Ludmila I. Kuncheva | 1 | 4942 | 244.34 |
Christopher J. Whitaker | 2 | 1242 | 55.85 |
Catherine A. Shipp | 3 | 255 | 11.34 |
R P W Duin | 4 | 662 | 50.98 |