세분화 기법을 이용한 데이터 통합에 관한 연구 = A Study on a Data Fusion using Segmentation Method

다국어 초록 (Multilingual Abstract)

This paper deals with data fusion, one of the ways to improve the quality of data, which is the source of information. Former studies on data fusion techniques have used the measure of similarity such as the distance between units. Considering that most of the ordinary data used not only in data mining but also in reality are large capacity ones, data fusion bears the burden that the amount of calculation(numbers of units, numbers of fusion variables) will be increased geometrically as the amount of data is increased.
To solve the problems mentioned above, Kim Soon-young and others(2005) suggested the clustering fusion technique. However, information available in this method will be limited, because it uses only the information of the continuous common variables in the clustering process. Therefore, the data fusion algorithm, which uses segmentation method as an alternative, is suggested in this paper.
The data fusion algorithm, using the segmentation method proposed in this paper, is a way to increase the effectiveness of fusion by making up for the shortcomings of previous methods, in which the amount of calculation and time increase when the researchers perform data fusion using large-capacity data. The units are segmented into groups of similar units, and then each segmented group performs the process of fusion. During this segmentation process, researchers can make segmentation groups using the variables which are shared in each file in order to minimize the uncertainty of fusion variables. When this method is used, it is possible to use all the information of common variables during the data segmentation process regardless of the metric of common variables. Researchers can not only choose and use common variables which play significant roles to the variables which are going to be fused, but also use the information of the unique variables that is supposed to be fused in order to minimize the uncertainty of fusion variables. To get this result, split algorithm, which is verified in the decision tree, was used to segment data.
If the most similar, one unit is used in the fusion process, other similar units can be ignored respectively. In this case, the accuracy of data fusion can be doubted. Therefore, several similar units, rather than the most similar one were used to perform data fusion.
The method suggested in this paper can not only reduce the amount of calculation and processing time for data fusion, but also lessen the loss of information of the respective, similar units. Therefore, it can increase the effectiveness of data fusion compared to the previous fusion algorithm.
In order to compare the effectiveness of the suggested segmentation technique with the previous one, an experiment was made with the real data. The suggested algorithm reduced the amount of calculations and time. It also improved the efficiency of data fusion in the respects of predicting power and representative characteristic of fusion files.

목차 (Table of Contents)

• 표목차 = ⅰ
• 그림목차 = ⅲ
• Abstract = ⅳ
• 제 1 장 서론 = 1
• 1.1 연구 배경 = 1
• 1.2 연구 목적 = 5
• 1.3 논문의 구성 = 8
• 제 2 장 데이터 통합 = 9
• 2.1 데이터 통합의 개념 = 9
• 2.2 데이터 통합의 종류 = 12
• 2.3 정확한 매칭과 통계적 매칭의 예시 = 17
• 제 3 장 통계적 매칭 = 19
• 3.1 통계적 매칭의 접근법 = 22
• 3.1.1 마이크로 접근법과 매크로 접근법 = 22
• 3.1.2 모수적 접근법과 비모수적 접근법 = 23
• 3.1.3 제약이 있는 매칭과 제약이 없는 매칭 = 24
• 3.2 일반적인 가정 = 26
• 3.2.1 조건부 독립 = 26
• 3.2.2 보조정보 이용 = 28
• 3.3 통계적 매칭의 타당성 = 30
• 3.4 통계적 매칭에서 상관계수의 불확실성 = 34
• 3.5 통계적 매칭의 수행과정 = 36
• 3.5.1 일반적인 절차 = 36
• 3.5.2 사전 전처리 과정 = 37
• 3.5.3 매칭변수 선택 = 39
• 3.5.4 근사성 측정 = 41
• 3.5.5 매칭 결과 평가 = 44
• 제 4 장 데이터 통합 알고리즘 = 46
• 4.1 기존의 통합 알고리즘 = 46
• 4.1.1 랜덤 핫덱 기법 = 46
• 4.1.2 순위 핫덱 기법 = 48
• 4.1.3 k-최근접이웃 기법 = 50
• 4.1.4 회귀분석 기법 = 53
• 4.1.5 회귀분석후 k-최근접이웃 기법 = 56
• 4.1.6 군집화 통합기법 = 58
• 4.2 세분화 기법을 이용한 데이터 통합 알고리즘 = 62
• 4.2.1 제안된 기법의 개념 = 62
• 4.2.2 데이터 세분화를 위한 의사결정나무의 분리 알고리즘 = 66
• 4.2.3 세분화 기법을 이용한 데이터 통합의 수행 단계 = 70
• 제 5 장 실증분석 = 74
• 5.1 UCI 데이터를 이용한 실증분석 = 76
• 5.1.1 데이터 설명 및 실험 절차 = 76
• 5.1.2 통계적 매칭의 평가 측도 = 79
• 5.1.3 통계적 매칭의 성능 평가 = 80
• 5.2 가계조사 데이터를 이용한 실증분석 = 87
• 5.2.1 데이터 설명 및 실험 절차 = 87
• 5.2.2 통계적 매칭 = 90
• 5.2.3 통계적 매칭의 성능 평가 = 94
• 5.2.4 데이터 통합 알고리즘의 수행시간 비교 = 104
• 5.3 가계조사와 가계자산조사를 이용한 통계적 매칭 = 106
• 5.3.1 데이터 설명 = 106
• 5.3.2 통계적 매칭 = 109
• 5.3.3 통계적 매칭의 결과 = 113
• 제 6 장 결론 및 향후 연구과제 = 119
• 참고문헌 = 122
• 부록 = 129