A0203
Title: Kernel association rotation analysis: A kernel-based projection of continuous data
Authors: Kimon Ntotsis - University of Leicester (United Kingdom) [presenting]
Andreas Artemiou - University of Limassol (Cyprus)
Alexandros Karagrigoriou - University of The Aegean (Greece)
Abstract: In high-dimensional data analysis, datasets often contain both linear and non-linear associations. Traditional dimensionality reduction techniques, like principal component analysis, focus primarily on linear associations, potentially missing important non-linear connections. This gap necessitates new methodologies capable of capturing the full spectrum of relationships. The aim is to present the kernel association rotation analysis (KARA), an unsupervised approach that excels in projecting the original data into a lower-dimensional space while retaining the majority of the original data's variance. KARA employs the kernel association coefficient (KAC), a measure designed to capture both linear and non-linear associations with high accuracy. Using a standardized 2-degree polynomial kernel function, KARA provides a comprehensive evaluation of feature relationships and data projection into a lower-dimensional space. Through simulations and real-world case studies, KARA's performance is evaluated and compared to other techniques using Gaussian process regression.
