Comparison of PCA with ICA from data distribution perspective

TL;DR

This paper empirically compares PCA and ICA algorithms on simulated noisy time series, highlighting ICA's superior performance due to its consideration of higher moments, while noting PCA's sensitivity to signal correlations.

Contribution

It provides an empirical analysis of PCA and ICA from a data distribution perspective, emphasizing differences in their robustness and effectiveness.

Findings

ICA outperforms PCA on noisy data

ICA considers higher moments of data distribution

PCA is sensitive to correlations among signals

Abstract

We performed an empirical comparison of ICA and PCA algorithms by applying them on two simulated noisy time series with varying distribution parameters and level of noise. In general, ICA shows better results than PCA because it takes into account higher moments of data distribution. On the other hand, PCA remains quite sensitive to the level of correlations among signals.