A new classification approach for handling new outcomes

Author

Li, Yingrong ; Kolesnikova, Anastasiya ; Don Lee, Won

Author_Institution

Dept. of Comput. Sci. & Eng., Chungnam Nat. Univ., Daejeon

fYear

2008

fDate

1-6 June 2008

Firstpage

3932

Lastpage

3937

Abstract

Classification is an important technique in the field of Data Mining and Machine Learning. The classifier can predict the class of unknown data based on their given attribute values. In ubiquitous computing environment, a great deal information can be obtained from various sensors. However, with the time going on, new sensor may be recruited. The recruited new sensors may bring new outcomes to the existing attribute. How to handle the new outcomes is a difficult issue. This paper first presents the problem and meanwhile a new method for handling new outcomes is proposed. The old rule is generated from the old data with fewer outcomes and modified and combined with the new data smoothly. In the method, the old rule can improve the performance of classifier constructed only from the new data set. The experiments show that the proposed approach is effective in handling new outcomes.

Keywords

data mining; learning (artificial intelligence); pattern classification; classification approach; data mining; machine learning; new outcome handling; ubiquitous computing; Classification algorithms; Computer science; Data mining; Decision trees; Machine learning; Recruitment; Sensor phenomena and characterization; Testing; Training data; Ubiquitous computing;

fLanguage

English

Publisher

ieee

Conference_Titel

Evolutionary Computation, 2008. CEC 2008. (IEEE World Congress on Computational Intelligence). IEEE Congress on

Conference_Location

Hong Kong

Print_ISBN

978-1-4244-1822-0

Electronic_ISBN

978-1-4244-1823-7

Type

conf

DOI

10.1109/CEC.2008.4631332

Filename

4631332