서브메뉴
검색
Cleaning and Learning Over Dirty Tabular Data
Cleaning and Learning Over Dirty Tabular Data
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105547
- ISBN
- 9798263395711
- DDC
- 004
- 저자명
- Li, Peng.
- 서명/저자
- Cleaning and Learning Over Dirty Tabular Data
- 발행사항
- [Sl] : Georgia Institute of Technology, 2023
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2023
- 형태사항
- 233 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
- 주기사항
- Advisor: Chu, Xu;Rong, Kexin.
- 학위논문주기
- Thesis (Ph.D.)--Georgia Institute of Technology, 2023.
- 초록/해제
- 요약The quality of Machine Learning (ML) applications is only as good as the quality of the data they train on. Unfortunately, real-world data is rarely free of errors, especially for tabular data, which frequently suffers from data issues like missing values, outliers, and inconsistencies. Therefore, data cleaning is widely regarded as an essential step in an ML workflow and an effective way to improve ML performance. However, data cleaning is often a time-consuming and expensive process that heavily relies on human efforts. Traditional data cleaning approaches often treat data cleaning as a standalone task independently of its downstream applications, which may not effectively improve ML performance and can sometimes even worsen it. Furthermore, it often leads to unnecessary costs for cleaning errors that have a minor impact on ML performance.This dissertation jointly considers data cleaning and machine learning, and focuses on developing algorithms and systems for cleaning and learning over dirty tabular data, with the dual objectives of (1) optimizing downstream ML performance and (2) minimizing human efforts.We start with a CLEANML study to empirically evaluate the impact of data cleaning on downstream ML performance. In this study, we build a benchmark that includes 14 real-world datasets with five common types of data errors, various data cleaning algorithms, and seven ML models. We conduct experiments using this benchmark and perform rigorous statistical testing to systematically investigate how data cleaning affects ML. This study establishes the feasibility of this dissertation and provides many insights for the following research.We then introduce CPCLEAN, a cost-effective human-involved data cleaning algorithm for ML that minimizes human cleaning efforts while preserving ML performance. The algorithm quantifies the impact of data cleaning on ML using a novel theoretical framework called Certain Predictions. It prioritizes cleaning examples with a large impact on ML while ignoring those with no impact. Our experiments show that CPCLEAN can reduce cleaning efforts by up to 80% without compromising ML performance.We subsequently demonstrate DIFFPREP, an automatic data cleaning method for ML that can automatically and efficiently select data preprocessing (cleaning) pipelines to maximize downstream ML performance without human involvement. We transform and relax the discrete, non-differential search space of data preprocessing pipelines into a continuous and differentiable one, which allows us to co-optimize data preprocessing and model training using gradient descent. Our experiments show that DIFFPREP achieves the best test accuracy on 15 out of the 18 datasets evaluated and improves the model's test accuracy by up to 6.6 percentage points.Finally, we present AUTO-TABLES that can automatically transform tables from nonstandard formats into a standard format without any human effort. This obviates the need for humans to manually program table-restructuring transformations for cleaning and learning over many real-world tables with non-standard formats, and broadens the application of our proposed cleaning methods to these tables. Our evaluation shows that AUTO-TABLES can successfully synthesize transformations for over 70% of test cases at interactive speeds.The works in this dissertation can be integrated into a comprehensive system for cleaning and learning over dirty tabular data. The dirty input tables will be first transformed into a standard format by AUTO-TABLES. Then, we manually clean data with CPCLEAN or automatically clean data with DIFFPREP. The system will minimize human efforts and output cleaned tabular data such that the downstream ML performance is optimized.
- 일반주제명
- Data science
- 기본자료저록
- Dissertations Abstracts International. 87-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2023 us c eng d■001000017360564
■00520260202105547
■006m o d
■007cr#unu||||||||
■020 ▼a9798263395711
■035 ▼a(MiAaPQ)AAI32315620
■035 ▼a(MiAaPQ)GeorgiaTech73152
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a004
■1001 ▼aLi, Peng.
■24510▼aCleaning and Learning Over Dirty Tabular Data
■260 ▼a[Sl]▼bGeorgia Institute of Technology▼c2023
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2023
■300 ▼a233 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: B.
■500 ▼aAdvisor: Chu, Xu;Rong, Kexin.
■5021 ▼aThesis (Ph.D.)--Georgia Institute of Technology, 2023.
■520 ▼aThe quality of Machine Learning (ML) applications is only as good as the quality of the data they train on. Unfortunately, real-world data is rarely free of errors, especially for tabular data, which frequently suffers from data issues like missing values, outliers, and inconsistencies. Therefore, data cleaning is widely regarded as an essential step in an ML workflow and an effective way to improve ML performance. However, data cleaning is often a time-consuming and expensive process that heavily relies on human efforts. Traditional data cleaning approaches often treat data cleaning as a standalone task independently of its downstream applications, which may not effectively improve ML performance and can sometimes even worsen it. Furthermore, it often leads to unnecessary costs for cleaning errors that have a minor impact on ML performance.This dissertation jointly considers data cleaning and machine learning, and focuses on developing algorithms and systems for cleaning and learning over dirty tabular data, with the dual objectives of (1) optimizing downstream ML performance and (2) minimizing human efforts.We start with a CLEANML study to empirically evaluate the impact of data cleaning on downstream ML performance. In this study, we build a benchmark that includes 14 real-world datasets with five common types of data errors, various data cleaning algorithms, and seven ML models. We conduct experiments using this benchmark and perform rigorous statistical testing to systematically investigate how data cleaning affects ML. This study establishes the feasibility of this dissertation and provides many insights for the following research.We then introduce CPCLEAN, a cost-effective human-involved data cleaning algorithm for ML that minimizes human cleaning efforts while preserving ML performance. The algorithm quantifies the impact of data cleaning on ML using a novel theoretical framework called Certain Predictions. It prioritizes cleaning examples with a large impact on ML while ignoring those with no impact. Our experiments show that CPCLEAN can reduce cleaning efforts by up to 80% without compromising ML performance.We subsequently demonstrate DIFFPREP, an automatic data cleaning method for ML that can automatically and efficiently select data preprocessing (cleaning) pipelines to maximize downstream ML performance without human involvement. We transform and relax the discrete, non-differential search space of data preprocessing pipelines into a continuous and differentiable one, which allows us to co-optimize data preprocessing and model training using gradient descent. Our experiments show that DIFFPREP achieves the best test accuracy on 15 out of the 18 datasets evaluated and improves the model's test accuracy by up to 6.6 percentage points.Finally, we present AUTO-TABLES that can automatically transform tables from nonstandard formats into a standard format without any human effort. This obviates the need for humans to manually program table-restructuring transformations for cleaning and learning over many real-world tables with non-standard formats, and broadens the application of our proposed cleaning methods to these tables. Our evaluation shows that AUTO-TABLES can successfully synthesize transformations for over 70% of test cases at interactive speeds.The works in this dissertation can be integrated into a comprehensive system for cleaning and learning over dirty tabular data. The dirty input tables will be first transformed into a standard format by AUTO-TABLES. Then, we manually clean data with CPCLEAN or automatically clean data with DIFFPREP. The system will minimize human efforts and output cleaned tabular data such that the downstream ML performance is optimized.
■590 ▼aSchool code: 0078.
■650 4▼aData science
■690 ▼a0800
■71020▼aGeorgia Institute of Technology.
■7730 ▼tDissertations Abstracts International▼g87-05B.
■790 ▼a0078
■791 ▼aPh.D.
■792 ▼a2023
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360564▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


