An Enhanced Projection Pursuit Method to Aid Pattern Recognition for Longitudinal Data

• Activity Number: 251
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Hua Fang and Zhaoyang Zhang*+ and Honggang Wang
• Companies: University of Massachusetts Medical School and University of Massachusetts Medical School/Dartmouth and University of Massachusetts/Dartmouth
• Keywords: Projection Pursuit ; High dimensional ; Longitudinal

Abstract

High-dimensional (HD) data are common in longitudinal studies. Projecting these complex data to a lower-dimensional space can tackle the problem of visualizing HD data. This can further help explore the data for patterns. However, HD data visualization techniques are still in improvement. We present an enhanced projection pursuit method that will improve the accuracy and stability of identified patterns, generally the structure of HD data. This method will automate the searching and find the optimal number of iterations to display a stable structure base on factors such as sample size and the number of variables. This method is compared with classical projection methods such as Andrew curves and dynamic grand tours using NIH-Funded longitudinal datasets for demonstration.