Pitch angle distributions (PADs) of suprathermal electrons serve as vital topological tracers of the interplanetary magnetic field, providing critical insights into the connectivity and evolution of solar wind transients. Traditional PAD analysis often relies on static, rule-based categorization into discrete shapes (e.g., strahl, halo, bidirectional), which struggles to capture the highly turbulent and transitional boundary layers of Interplanetary Coronal Mass Ejections (ICMEs) and Stream Interaction Regions (SIRs).

In this study, we present a robust, probabilistic machine learning framework designed to capture PAD evolution. Utilizing 15 years of solar wind data from the Advanced Composition Explorer (ACE), we first applied a Fuzzy C-Means (FCM) clustering algorithm to identify 32 pure geometric archetypes. These clusters were then mapped to nine broader physical categories (e.g., Narrow Strahl, Scattered CSE, Pancake) to train a highly generalized Random Forest classifier.

We demonstrate the application of this trained model across multiple observational platforms, applying it to both the full 15-year ACE dataset and 8 years of data from the STEREO-A and STEREO-B observatories. By outputting multi-class probabilities rather than single discrete labels, this allows for the continuous visualization of PAD evolution—highlighting dual-membership states where spacecraft encounters more than one suprathermal electron population at a time (strahl and pancake). We present multi-spacecraft case studies showcasing how this tool traces the simultaneous topological evolution of transients across different heliospheric coordinates.

Furthermore, we discuss the pipeline’s potential for large-scale occurrence rate analysis and its upcoming application to Parker Solar Probe (PSP) data. We invite discussion on how probabilistic ML outputs can be better integrated into global heliospheric models.