講   題: 多重時間序列資料之維度縮減與視覺化:象徵性資料分析方法

Dimension reduction and visualization of multiple time series data: a symbolic data analysis approach

主 講 人:台北大學統計系 吳漢銘 副教授 (Han-Ming Wu)

Department of Statistics, National Taipei University

主 持 人:王志軒 教授

主辦單位:交通大學工業工程與管理系

時   間: 108年 10月 7日(星期一) 13:20 ~ 15:20

地   點:管二館520室

演講摘要:

Exploratory analysis and visualization, which are performed prior to the modeling and forecasting of multiple time series data, present  important steps in the discovery of the underlying dynamics of a series. In  this study, we present extensions of two dimension reduction (DR) methods,  principal component analysis (PCA) and sliced inverse regression (SIR), to  multiple time series data through a symbolic data analysis framework, named the  path point approach. First, the multiple time series data are put in the form  of time-dependent intervals such that the interval is described by a starting  value and an ending value of a certain time period. In this context, each  series can be geometrically represented as consecutive directed segments with path points. Then, we apply PCA and SIR to the data table formed by the coordinates of these  path points to visualize the temporal trajectories of objects in a lower dimensional subspace. Experimental results from studies of a simulation, microarray time series data concerning a yeast cell cycle and the financial data show that the proposed methods are capable of providing insight into the structure and behaviors of objects on the 2D factorial axes. Comparisons with the currently existing method, namely, the applied vector  approach, for PCA and SIR for time-dependent interval data are also examined.

演講性質:學術研究專題

歡迎聽講