UMP Institutional Repository

Evolutionary-based feature construction with substitution for data summarization using DARA

Sia, Florence and Alfred, Rayner (2012) Evolutionary-based feature construction with substitution for data summarization using DARA. In: IEEE 4th Conference on Data Mining and Optimization (DMO 2012), 2-4 September 2012 , Langkawi, Kedah. pp. 53-58.. ISBN 978-1-4673-2718-3

[img]
Preview
Pdf
Evolutionary-based feature construction with substitution for data.pdf

Download (274kB) | Preview

Abstract

The representation of input data set is important for learning task. In data summarization, the representation of the multi-instances stored in non-target tables that have many-to-one relationship with record stored in target table influences the descriptive accuracy of the summarized data. If the summarized data is fed into a classifier as one of the input features, the predictive accuracy of the classifier will also be affected. This paper proposes an evolutionary-based feature construction approach namely Fixed-Length Feature Construction with Substitution (FLFCWS) to address the problem by means of optimizing the feature construction for relational data summarization. This approach allows initial features to be used more than once in constructing newly constructed features. This is performed in order to exploit all possible interactions among attributes which involves an application of genetic algorithm to find a relevant set of features. The constructed features will be used to generate relevant patterns that characterize non-target records associated to the target record as an input representation for data summarization process. Several feature scoring measures are used as fitness function to find the best set of constructed features. The experimental results show that there is an improvement of predictive accuracy for classifying data summarized based on FLFCWS approach which indirectly improves the descriptive accuracy of the summarized data. It shows that FLFCWS approach can generate promising set of constructed features to describe the characteristics of non-target records for data summarization.

Item Type: Conference or Workshop Item (Lecture)
Additional Information: Indexed by Scopus
Uncontrolled Keywords: relational data mining; relational data summarization; feature construction; genetic algorithm
Subjects: Q Science > QA Mathematics > QA76 Computer software
Faculty/Division: Faculty of Computer System And Software Engineering
Institute of Postgraduate Studies
Depositing User: Mrs. Neng Sury Sulaiman
Date Deposited: 22 Mar 2020 23:30
Last Modified: 22 Mar 2020 23:30
URI: http://umpir.ump.edu.my/id/eprint/26997
Download Statistic: View Download Statistics

Actions (login required)

View Item View Item