Feature Selection Based on the Training Set Manipulation - PhD thesis proposal

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F05%3A00109895" target="_blank" >RIV/68407700:21230/05:00109895 - isvavai.cz</a>

Výsledek na webu

—

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Feature Selection Based on the Training Set Manipulation - PhD thesis proposal

Popis výsledku v původním jazyce

A novel feature selection technique for the classification problems is proposed in this PhD thesis proposal. The method is based on the training set manipulation. A weight is associated with each training sample similarly as it is in the AdaBoost algorithm. The weights form a distribution. Any change of the distribution of weights influences the behaviour of particular features in a different manner. This brings new information to the selection process in contrast to other feature selection techniques.The main idea is to modify the weights in each selection step so that the currently selected feature appears, with respect to the distribution, like an irrelevant observation. We show in experiments that such a change of the weights distribution allows to reveal hidden relationships between features. Although the feature selection algorithm is not completely developed yet, preliminary results achieved on several artificial problem looks promising.

Název v anglickém jazyce

Feature Selection Based on the Training Set Manipulation - PhD thesis proposal

Popis výsledku anglicky

A novel feature selection technique for the classification problems is proposed in this PhD thesis proposal. The method is based on the training set manipulation. A weight is associated with each training sample similarly as it is in the AdaBoost algorithm. The weights form a distribution. Any change of the distribution of weights influences the behaviour of particular features in a different manner. This brings new information to the selection process in contrast to other feature selection techniques.The main idea is to modify the weights in each selection step so that the currently selected feature appears, with respect to the distribution, like an irrelevant observation. We show in experiments that such a change of the weights distribution allows to reveal hidden relationships between features. Although the feature selection algorithm is not completely developed yet, preliminary results achieved on several artificial problem looks promising.

Druh

O - Ostatní výsledky

CEP obor

JD - Využití počítačů, robotika a její aplikace

OECD FORD obor

—

Projekt

<a href="/cs/project/GA102%2F03%2F0440" target="_blank" >GA102/03/0440: Rozpoznávání lidských aktivit pro automatické sledování z videa</a><br>

Návaznosti

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

Rok uplatnění

2005

Kód důvěrnosti údajů

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