개인구독
소속 기관이 없으신 경우, 개인 정기구독을 하시면 저렴하게
논문을 무제한 열람 이용할 수 있어요.
지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
논문 기본 정보
- 자료유형
- 학위논문
- 저자정보
- 지도교수
- 김현철
- 발행연도
- 2020
- 저작권
- 고려대학교 논문은 저작권에 의해 보호받습니다.
이용수9
이 논문의 연구 히스토리 (2)
- 이 논문의 후속연구가 궁금하신가요?
- 연관 학술논문 또는 학술발표를 통해 보다 발전된 연구결과를 확인하실 수 있습니다.
- 이 논문의 연구 히스토리 확인하기
초록· 키워드
오류제보하기
데이터 불균형은 머신러닝 분류 모델의 성능을 저하하는 요인으로 작용한다고 알려져 있다. 이러한 불균형을 해결하기 위해 샘플링 방법이나 두 개 이상의 모델을 결합한 앙상블 기법 등을 이용하여 분류하려는 집단의 데이터 분포를 비슷하게 맞추고 성능을 높이려는 연구가 활발하게 진행되고 있다.
본 연구에서는 데이터의 불균형이 머신러닝 모델 성능에 미치는 영향을 알아보기 위해 역으로 클래스에 데이터의 불균형을 만들어 실험하였다. 실험에 사용된 데이터는 MNIST 손글씨 숫자 이미지 데이터이며, 머신러닝 모델은 총 4개로 컨볼루션 신경망(Convolutional Neural Network; CNN), 의사결정 나무(Decision Tree; DT), 로지스틱 회귀(Logistic Regression; LR), k-최근접 이웃(k-Nearest Neighbor; k-NN)이다. 실험은 세 가지 기준(클래스 0~4가 불균형일 때, 특정 클래스가 불균형일 때, 전체 클래스 데이터양에 차이가 있을 때)으로 데이터를 각각 10% 단위로 삭제한 후 실험하였다. 그중 데이터 불균형이 10, 40, 80%일 때를 선정하여 분석하였고 실제로 불균형이 클수록 머신러닝 모델 성능에 영향이 있음을 확인할 수 있었다.
본 연구에서는 데이터의 불균형이 머신러닝 모델 성능에 미치는 영향을 알아보기 위해 역으로 클래스에 데이터의 불균형을 만들어 실험하였다. 실험에 사용된 데이터는 MNIST 손글씨 숫자 이미지 데이터이며, 머신러닝 모델은 총 4개로 컨볼루션 신경망(Convolutional Neural Network; CNN), 의사결정 나무(Decision Tree; DT), 로지스틱 회귀(Logistic Regression; LR), k-최근접 이웃(k-Nearest Neighbor; k-NN)이다. 실험은 세 가지 기준(클래스 0~4가 불균형일 때, 특정 클래스가 불균형일 때, 전체 클래스 데이터양에 차이가 있을 때)으로 데이터를 각각 10% 단위로 삭제한 후 실험하였다. 그중 데이터 불균형이 10, 40, 80%일 때를 선정하여 분석하였고 실제로 불균형이 클수록 머신러닝 모델 성능에 영향이 있음을 확인할 수 있었다.
목차
제 1 장 서 론 ········································································· 11.1 연구배경 및 목적 ··························································· 11.2 논문의 구성 ·································································· 2제 2 장 관련 연구 ·································································· 2제 3 장 연구 계획 ··································································· 33.1 실험 방법 ······································································ 33.2 데이터 준비 ··································································· 3제 4 장 연구 결과 ··································································· 54.1 실험 1: 클래스 0∼4가 불균형일 때 ································· 54.1.1 컨볼루션 신경망 ························································· 54.1.2 의사결정 나무 ···························································· 74.1.3 로지스틱 회귀 ···························································· 94.1.4 k-최근접 이웃 ··························································· 114.1.5 실험 1: 종합 결과 ······················································ 134.2 실험 2: 특정 클래스(1, 3, 4, 6, 7)가 불균형일 때 ············ 154.2.1 컨볼루션 신경망 ························································ 164.2.2 의사결정 나무 ··························································· 194.2.3 로지스틱 회귀 ··························································· 224.2.4 k-최근접 이웃 ··························································· 254.2.5 실험 2: 종합 결과 ······················································ 284.3 실험 3: 전체 클래스의 데이터양에 차이가 있을 때 ············ 294.3.1 컨볼루션 신경망 ························································ 294.3.2 의사결정 나무 ··························································· 324.3.3 로지스틱 회귀 ··························································· 354.3.4 k-최근접 이웃 ··························································· 384.3.5 실험 3: 종합 결과 ······················································ 41제 5 장 결론 ·········································································· 44참고문헌 ··············································································· 45
최근 본 자료
댓글(0)
0