Comparison of dimension reduction techniques applied to the analysis of airborne radionuclide activity concentration

Abstract

Temporal variation of airborne radionuclide activity concentration is highly influenced by several meteorological parameters. A traditional key tool to perform a combined analysis on these data is Principal Component Analysis (PCA), a linear dimensionality reduction technique that prioritizes the conservation of the data set global structure. While it reveals important information regarding the correlation among the considered variables, the obtained visual representations do not usually allow to clearly discern different clusters of states with common properties. The main goal of this study is applying two recently introduced non linear dimensionality reduction techniques, t-distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP) to a data set composed by 7Be and gross beta (Aβ) activity concentration and other meteorological data gathered in Mallorca (Spain) between 2004 and 2014. Compared to PCA, both algorithms reveal more details on the local structure of the data set. UMAP allows to clearly identify data clusters with different characteristics that are not clearly identified with the alternative techniques.