의사결정나무 분류 모델 해석력 향상을 위한 새로운 분기 기준 :A New Branching Criterion for Improving Comprehensibility of Decision Tree Classification Model

여현규

추천

검색

자료유형: 학위논문

저자정보: 여현규 (서울과학기술대학교, 서울과학기술대학교 대학원)

지도교수: 홍정식

발행연도: 2018

저작권: 서울과학기술대학교 논문은 저작권에 의해 보호받습니다.

이용수1

이 논문의 연구 히스토리 (2)

2018

의사결정나무 분류 모델 해석력 향상을 위한 새로운 분기 기준

여현규 , 홍정식 , 황상흠 한국경영과학회 학술대회논문집 2018.10 학술대회자료

의사결정나무 분류 모델 해석력 향상을 위한 새로운 분기 기준

여현규 데이터사이언스학과 2018.01 학위논문

이 논문의 후속연구가 궁금하신가요?
연관 학술논문 또는 학술발표를 통해 보다 발전된 연구결과를 확인하실 수 있습니다.
이 논문의 연구 히스토리 확인하기

초록· 키워드

오류제보하기

의사결정나무는 입력변수의 차원으로 이루어진 데이터 공간을 재귀적으로 분할하는 과정에서 하나의 규칙을 생성하며, 각각의 규칙들은 계층적으로 표현된다. 계층적으로 표현된 규칙들은 분석가에게 목표변수와의 관계를 파악할 수 있는 정보를 제공한다. 그러나 이러한 구조의 규칙은 단순할 때는 직관적인 정보를 제공하지만, 복잡해지면 그것을 이해하는 것이 매우 어렵다. 또한 각각의 규칙에 해당하는 샘플의 수가 너무 적거나, 동질성이 낮다면 이 또한 분석가에게 충분한 정보를 제공한다고 보기 어렵다. 이에 본 논문에서는 해석력을 높이기 위한 새로운 분기 기준을 제안한다. 제안한 분기 기준에 대해 특성을 분석하고, 검증을 위해 성능 측면과 해석력 측면에서 기존 모델과의 비교를 수행한다. 기존 모델과의 비교에는 총 21가지 데이터 셋이 사용되었으며 성능의 경우 정확도(Accuracy), 민감도(Sensitivity, Recall), 정밀도(Precision), F1- measure, AUC(Area Under the Curve)를 성능 측정 지표로 활용하였다. 성능 측정 결과 AUC(macro)를 기준으로 제안한 모델은 11개의 사례에서 높았고, CART(Classification And Regression Trees)는 2개의 사례에서 높았다.(유의수준 1%) 해석력의 경우 성능 측정 과정에서 생성된 모델 중 구조적으로 다른 몇 가지 예시를 확인하고, 수치적인 비교를 수행하였다. 생성된 예시를 통해 제안한 모델이 CART 보다 동질성 높은 규칙을 생성하면서 One-sided tree 보다 많은 샘플을 포함하는 것을 확인하였다. 해석력에 대한 수치는 깊이가 얕은 단순한 규칙에 포함 되는 샘플의 비율과 목표변수 범주의 최대 확률 값을 통해 측정하였다. 그 결과 깊이가 3 이하에서 샘플 비율과 동질성이 모두 높은 사례는 제안한 모델이 6개, CART가 1개로 제안한 모델이 5개 더 많았으며, 깊이가 4이하인 경우에서도 제안한 모델이 4개, CART가 1개로 제안한 모델이 높은 사례가 더 많았다.(유의수준 1%) 비교 실험을 통해 우리가 제안한 분기 기준이 성능을 더 높이면서도 해석력 좋은 규칙을 유도하는 것을 확인하였다.

Decision trees generate a rule in the process of recursively dividing the data space consisting of the dimensions of the input variables, and each rule is expressed hierarchically. Hierarchically represented rules provide the analysts with information that can identify the relationship with the target variables. The rules of hierarchical structure provide intuitive information when simple, but it is very difficult to understand if it becomes complicated. Also, if the number of samples for each rule is too small or homogeneity is low, it is difficult to say that the rule provides sufficient information. In this paper, we propose a new branching criterion to improve comprehensibility. We analyze the properties of the proposed branching criterion and compare it with the existing model in terms of performance and comprehensibility. A total of 21 data sets were used in comparison with the existing model. In the case of performance, accuracy, sensitivity, recall, precision, F1-measure and AUC(Area Under the Curve) were used as performance measures. The proposed model performance based on the AUC (macro) was higher in 11 cases and the CART (Classification And Regression Trees) was higher in 2 cases. (significance level 1%) In the case of comprehensibility. Several generated examples were identified and numerical comparisons were performed. Through the generated examples, it is confirmed that the proposed model includes more samples than the one-sided tree while generating the homogeneous rule than CART. The numerical value of the comprehensibility was measured by the ratio of the samples included in the shallow depth rule and the maximum probability value of the target variable category(homogeneity). As a result, in the case of the simple rule with depth of 3 or less, the proposed model was higher in 6 data, CART was higher in 1 data. In the case of a depth of less than 4, the proposed model was higher in 4 data, CART was higher in 1 data.(significance level 1%). Comparative experiments have shown that the branching criterion we propose lead to better performance and better interpretable rules.

1. 서 론 ·····························································································································1
2. 관련연구 ·······················································································································3
2.1. 분기 기준 및 알고리즘에 대한 기존 연구 ························································3
2.2. 해석력에 대한 기존 연구 ······················································································5
3. 새로운 분기 기준 ·······································································································6
3.1. One-sided tree ·········································································································6
3.2. 새로운 분기 기준 ····································································································7
3.3. 새로운 분기 기준의 특성 ··················································································10
4. 실 험 ···························································································································18
4.1. 실험 데이터 ············································································································18
4.2. 해석력 ······················································································································19
4.3. 성능 ··························································································································25
5. 결 론 ···························································································································30
참고문헌 ··························································································································32
부 록 ································································································································36
영문초록(Abstract) ·········································································································39

최근 본 자료

전체보기

구분	그룹	데이터 항목
AI 학습용 데이터	원문	원문 PDF 파일
AI 학습용 데이터	원문 + 메타 (기본/상세)	원문 PDF 파일 및 서지정보 CSV
대량 구매용 데이터	B2B 구독 방식	특정 자료 한정으로 원문 접근 권한 부여
대량 구매용 데이터	URL 전달 방식	바로 PDF 뷰어를 열람할 수 있는 URL 제공

구분	그룹	데이터 항목
AI 학습용 데이터	기본 메타	발행기관명, 간행물명, 권호명, 권(vol), 호(issue), 통권, 발행연도, 발행월, 논문명, 저자명, 시작페이지, 종료페이지, 전체페이지, 상세페이지URL
상세 메타 데이터	발행기관 메타	발행기관 이명, 영문명, 창립연도, 홈페이지URL, 발행기관 소개
	간행물 메타	부제목, 간행물 유형, ISSN, ISBN, 최초발행연도, 폐간연도, 간행빈도, 발행주기, 등재사항, 이용수, 피인용수, 권호수, 논문수, 표지이미지
	논문 메타	작성 언어, 부제목, 대등제목, 목차, 키워드, 초록, 이미지, 참고문헌, 이용수, 피인용수, 논문활용도, DBpia통합주제분류, KDC분류, DDC분류, 한국연구재단분류, UCI, DOI
	저자 메타	소속기관, 소속부서, 직급, 연구분야, 연구키워드, 이용수, 피인용수, 저자 논문활용도

구분	그룹	데이터 항목
※ 결합형/맞춤형 메타 데이터는 신청 내용에 따라 다양하게 제공 가능
이용순위 정보	주제분야별 많이 이용된 논문	“인문학”에서 많이 이용된 논문 TOP100
	이용기관별 많이 이용된 논문	“중고등학교”에서 많이 이용된 논문 TOP100
	세부기관별 많이 이용된 논문	“서울대학교”에서 많이 이용된 논문 TOP100
	키워드별 많이 이용된 논문	“Chat GPT”에서 많이 이용된 논문 TOP100
키워드 정보	많이 이용된 키워드	특정기간/분야/저널 내 많이 이용된 키워드
	많이 발행된 키워드	특정기간/분야/저널 내 많이 발행된 키워드
	많이 검색된 키워드	특정기간/분야/저널 내 많이 검색된 키워드
	연구 트렌드 키워드	특정 키워드 연관 연구동향 분석 데이터 키워드

논문 기본 정보

이 논문의 연구 히스토리 (2)

초록· 키워드

목차

최근 본 자료

댓글(0)