Characterizing Suprathermal Electron Propagation Through Pitch Angle Distributions in the Inner Heliosphere

Suprathermal electron pitch-angle distributions (PADs) are sensitive diagnostics of magnetic connectivity, scattering, and large-scale heliospheric structure. While these distributions have been studied for decades, most classification approaches rely on manual inspection or physics-specific heuristics. Here we present the first fully data-driven framework that applies both unsupervised and supervised machine learning technqiues directly to raw PAD fluxes—without smoothing, hand-engineered features, or embedded physical assumptions. Using Advanced Composition Explorer (ACE) observations, an unsupervised K-Means algorithm operating solely on normalized PADs, recovers many canonical PAD morphologies reported in the literature, including narrow and broad strahl, counterstreaming, loss-cone, pancake, and butterfly type distributions. These emergent clusters provide a physically meaningful basis for training a random forest classifier that generalizes across a full solar cycle and transfers robustly to STEREO. The combined pipeline is instrument-agnostic, computationally efficient, and capable of identifying less common morphologies such as butterfly and pancake PADs, which we find to recur during the most intense Earth-directed ICME events. Their presence suggests retained magnetic structure or partial field-line coherence within CME interiors, highlighting an underutilized diagnostic potential of suprathermal electrons for studying ICME topology and space weather geoeffectiveness. This work establishes a scalable, reproducible framework for PAD classification and opens new directions for leveraging data-driven discovery in heliospheric and planetary space environments, including near-lunar and Martian regimes where wave–particle interactions shape suprathermal electron behavior.