Data Dimensionality Reduction Based on the Ica Approach

Dimensionality reduction represents an important task in the field of statistical pattern recognition and machine learning in general. One of the most commonly used methods for this purpose is Principal Component Analysis (PCA), which appears under various names in the literature. Its primary goal is to decorrelate the components within the space of newly obtained coordinates. The approach that does not consider decorrelation sufficient, but instead requires that the components in the transformed space be statistically independent, is known as Independent Component Analysis (ICA). In this way, dimensionality reduction is achieved with minimal information loss caused by discarding original coordinates. An analysis of existing literature reveals that the application of this method is highly sensitive not only to the available training dataset, but also to the statistical distribution of data vectors. This paper analyzes the capabilities and limitations of such an approach to dimensionality reduction.