Interactive Visual Parameter Exploration in Functional Data Analysis
Truls H. Jakobsen${}^a$​,   Ana-Maria Urdea${}^b$​,   Rainer Splechtna${}^c$​,   Krešimir Matković${}^c$​,   Peter Filzmoser${}^b$​,   Helwig Hauser${}^a$
${}^a$Univ. of Bergen, Norway;  ${}^b$TU Wien, Austria;  ${}^c$VRVis Research Center, Austria
Functional data analysis (FDA) treats observations as functions rather than as isolated measurements Ramsay and Silverman [2005], and a central challenge in any FDA workflow is the configuration of the function representation itself. Analysts must choose a set of basis functions, determine the appropriate parameterization of that basis (number of basis functions, order, knot placement, and so on), and select an adequate smoothing level before they can inspect the consequences of their choices in any detail. In script-driven environments, revisiting these decisions requires rewriting and rerunning code, making it difficult to develop intuition about model sensitivity or to compare alternatives efficiently.

We present a prototype that addresses this by connecting R directly to an interactive visualization application. We integrate ComVis, a coordinated multiple views framework for prototyping visualization technology Matković et al. [2008], with the fda package Ramsay and Silverman [2005] as a statistical backend.

The result is a semi-guided workflow in which parameter choices are exposed as interactive controls, and linked views provide immediate visual feedback on how each choice affects the fitted functions and their derivatives. Analysts can brush subsets of interest and observe modelling consequences without leaving the visual environment. Immediate visual feedback guides the analyst toward an appropriate FDA pipeline configuration, and the ability to display multiple parameter configurations simultaneously provides a direct means of comparison and a clearer understanding of parameter influence.

Keywords: Functional data analysis, interactive visualization, visual analytics