Visualization and Interpretation of Unsupervised Solar Wind Classifications

Abstract:

One of the goals of machine learning is to eliminate tedious and arduous repetitive work. The manual classification of millions of hours of solar wind data from multiple missions can be replaced by automatic algorithms that can discover the real differences in the solar wind properties. Here we present how unsupervised clustering techniques can be used to segregate different types of solar wind. We propose the use of advanced data reduction methods to pre-process the data, and we introduce the use of multiple unsupervised classification methods to visualize and interpret 14 years of ACE data. Finally, we show how these techniques can potentially be used to uncover hidden information, and how they compare with previous empirical categorizations.

We use two main types of data pre-processing: Kernel PCA and Autoencoders. This data transformation step allows to project the original data in a more meaningful latent space. The unsupervised classification and the visualization of the data is performed using multiple methods for comparison, including k-means and Bayesian Gaussian Mixtures. We introduce and promote the use of Dynamic Self-Organizing Maps to cluster and visualize the complex multi-dimensional data from ACE. All these visualization and clustering techniques are complementary, and their results still require a skeptical interpretation.

The figure attached shows the general overview of the pipelines tested in this work. Starting from the center, the ACE data set is processed and normalized. Blue dashed lines show the work done in previous publications by different authors. Black lines show how data in this work is first transformed and then classified using multiple methods. The original techniques presented in this work are highlighted in red.

The work presented here has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 776262 (AIDA, www.aida-space.eu).