Dynamic Time Warping based t-SNE for Trajectory data : : Two Real-Data Applications
- 주제(키워드) Dynamic time warping , t-distributed stochastic neighbor embedding , trajectory data , visualization
- 주제(DDC) 510
- 발행기관 아주대학교 일반대학원
- 지도교수 안수현
- 발행년도 2025
- 학위수여년월 2025. 2
- 학위명 석사
- 학과 및 전공 일반대학원 수학과
- 실제URI http://www.dcollection.net/handler/ajou/000000034289
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
In this study, we focus on visualizing trajectory data, a type of data based on a series of temporal observations. We present an adapted version of t-distributed Stochastic Neighbor Embedding (t- SNE) tailored for trajectory data. This method is designed to preserve the inherent curved structure of trajectory data by incorporating a robust distance measure. Furthermore, it demonstrates the ability to maintain the data structure even in the presence of missing values at different time points. In exploring suitable similarity metrics for this task, we investigated four different similarity metrics and ultimately selected Dynamic Time Warping (DTW) as the most appropriate for capturing the temporal structure of the data. The performance of the proposed method is rigorously evaluated through a simulation study, demonstrating its effectiveness in visualizing two types of trajectory data: Gait data and NBA data.
more목차
1. Introduction 1
2. Similarity Metrics 3
2.1 Data and Notation 3
2.2 Dynamic Time Warping 4
2.3 Fréchet distance 5
2.4 String Metrics 6
2.4.1 Longest Common Subsequence (LCS) 6
2.4.2 Edit distance 6
3. Methods 8
3.1 t-distributed Stochastic Neighbor Embedding 8
3.2 DTW based t-SNE 9
4. Numerical Study 13
5. Two Real-Data Applications 17
5.1 Gait Data 17
5.2 NBA Data 22
6. Conclusion 25
Reference 29
