Approaches to Functional Time Series Data Analysis With Applications to Physical Activity Measurements- [electronic resource]
Approaches to Functional Time Series Data Analysis With Applications to Physical Activity Measurements- [electronic resource]
- 자료유형
- 학위논문파일 국외
- 최종처리일시
- 20240214101925
- ISBN
- 9798380848220
- DDC
- 614.4
- 서명/저자
- Approaches to Functional Time Series Data Analysis With Applications to Physical Activity Measurements - [electronic resource]
- 발행사항
- [S.l.]: : Indiana University., 2023
- 발행사항
- Ann Arbor : : ProQuest Dissertations & Theses,, 2023
- 형태사항
- 1 online resource(149 p.)
- 주기사항
- Source: Dissertations Abstracts International, Volume: 85-05, Section: B.
- 주기사항
- Advisor: Harezlak, Jaroslaw.
- 학위논문주기
- Thesis (Ph.D.)--Indiana University, 2023.
- 사용제한주기
- This item must not be sold to any third party vendors.
- 초록/해제
- 요약The central topic in the three parts of my dissertation is the use of accelerometer-derived data in public health applications. In parts one and two, we examine real-world and simulated accelerometry data using functional data analysis techniques. We show that the multilevel functional principal component analysis (MFPCA) approach, originally developed to account for a common-component correlation, can be meaningfully applied in settings featuring more complex correlation patterns. In part one, we evaluate several procedures for selecting the number of functional principal components to retain after MFPCA is applied to a simulated collection of functional time series. The data are generated from a modified functional linear mixed effects model that allows for dependencies within and across subjects. A real-world equivalent to this structure could be data collected from players of a team sport. Realistic sample sizes in the lower 100s can provide accurate recovery of the shape and number of components on the player-specific level under both dependence types. While we find no uniformly optimal method, the most intuitive ones relying on the idea of a scree plot perform most reliably. Parts two and three use preprocessed real-world accelerometer-derived walking data from a multi-subject study. These data are transformed into the frequency domain to obtain walking spectra. In part two, we apply structured functional principal component analysis (SFPCA) to extract features from walking signals on the subject and the subject-spectrum level. In regressions, we associate the subject-level feature scores with age and physical health measures. SFPCA decomposes these spectra into easily interpretable features, which can be related to physical performance, potentially shedding light on subclinical disease. In part three, we propose a gait-based algorithm for the recognition of individuals. The manner of walking is primarily determined by physical features and individual-specific behaviors and can thus be used as a biometric identifier. Different from existing procedures, our method does not require any supplemental algorithms once the data is transformed into the frequency domain. Instead, we correlate the spectra within and across individuals and find that intra-subject curves are more strongly correlated than inter-subject curves. Thus, our algorithm can accurately identify the participants.
- 일반주제명
- Epidemiology.
- 일반주제명
- Biostatistics.
- 일반주제명
- Public health.
- 키워드
- Time series
- 키워드
- Walking data
- 기타저자
- Indiana University School of Public Health
- 기본자료저록
- Dissertations Abstracts International. 85-05B.
- 기본자료저록
- Dissertation Abstract International
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.