Empirically Comparing Two Dimensionality Reduction Techniques – PCA and FFT: A Settlement Detection Case Study in the Gauteng Province of South Africa

In this paper we present a class label agnostic dimensionality reduction comparison framework. We illustrate the usefulness of this framework at the hand of a case study. For our case study, we consider two prominent land cover classes in the Gauteng province, namely natural vegetation and settlement using an 8 year MODIS dataset. We use the framework to compare two feature extraction techniques, namely PCA and FFT. For the case study we considered in this paper, the PCA technique produced a reduced feature space which was 15% more separable than the feature space produced by the FFT method.