Distribution of a Non-parametric Wavelet-based Statistic for Functional Data

TL;DR

This paper provides a mathematical foundation for the distribution of a wavelet-based statistic used in functional data analysis, assuming normality, and discusses potential extensions to non-Gaussian data.

Contribution

It develops a rigorous distributional theory for a wavelet-based statistic with hard thresholding in high-dimensional functional data analysis.

Findings

Analytic distribution derived under normality assumption

Proposes approximations for the wavelet-based statistic

Framework applicable to high-dimensional data analysis

Abstract

Mathematical formulations and proofs for a wavelet based statistic employed in functional data analysis is elaborately discussed in this report. The propositions and derivations discussed here apply to a wavelet based statistic with hard thresholding. The proposed analytic distribution is made feasible only due to the assumption of normality. Since the statistic is developed for applications in high dimensional data analysis, the assumption holds true in most practical situations. In the future, the work here could be extended to address data that are non-Gaussian. Aside from establishing a rigorous mathematical foundation for the distribution of the statistic, the report also explores a few approximations for the proposed statistic.