Experimentální studie vážení listů pro metodu Random Forests

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F67985807%3A_____%2F04%3A00105191" target="_blank" >RIV/67985807:_____/04:00105191 - isvavai.cz</a>

Výsledek na webu

—

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Experimental Study of Leaf Confidences for Random Forest

Popis výsledku v původním jazyce

Decision forests (ensembles of trees) achieve usually smaller generalization error compared to single trees. In the classical methods for growing forests, bagging and boosting, the individual trees are constructed by methods originally developed for growing a single tree as the final predictor. In particular, the trees are usually pruned. For such trees, using weights (confidences) for individual trees improves the accuracy of the prediction of the ensemble. Random forests technique (Breiman 2001) usesa specific tree growing process, which does not produce good individual trees, but the whole ensemble frequently achieves better results than ensembles of trees obtained by classical bagging and boosting. One of the default features of Random Forests technique is that it does not use any weights. The current paper presents experiments demonstrating that in specific situations, appropriately chosen weights may improve the prediction for Random Forests of limited size.

Název v anglickém jazyce

Experimental Study of Leaf Confidences for Random Forest

Popis výsledku anglicky

Decision forests (ensembles of trees) achieve usually smaller generalization error compared to single trees. In the classical methods for growing forests, bagging and boosting, the individual trees are constructed by methods originally developed for growing a single tree as the final predictor. In particular, the trees are usually pruned. For such trees, using weights (confidences) for individual trees improves the accuracy of the prediction of the ensemble. Random forests technique (Breiman 2001) usesa specific tree growing process, which does not produce good individual trees, but the whole ensemble frequently achieves better results than ensembles of trees obtained by classical bagging and boosting. One of the default features of Random Forests technique is that it does not use any weights. The current paper presents experiments demonstrating that in specific situations, appropriately chosen weights may improve the prediction for Random Forests of limited size.

Druh

D - Stať ve sborníku

CEP obor

BA - Obecná matematika

OECD FORD obor

—

Projekt

<a href="/cs/project/LN00A056" target="_blank" >LN00A056: Institut teoretické informatiky - Centrum mladé vědy</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2004

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Computational Statistics

ISBN

3-7908-1554-3

ISSN

—

e-ISSN

—

Počet stran výsledku

8

Strana od-do

1767-1774

Název nakladatele

Physica Verlag

Místo vydání

Heidelberg

Místo konání akce

Prague

Datum konání akce

23. 8. 2004

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—