Identifying group galaxies merging with massive clusters using machine learning

The environment plays a critical role in galaxy evolution, with galaxy clusters and their infall regions offering diverse conditions that shape galaxies before they enter the dense cluster core, a process known as ``pre-processing''. However, identifying environmental substructures, particularly galaxy groups in these transitional zones, remains challenging due to projection effects and ``fingers-of-god'' distortions. In this work, we present a supervised machine learning framework for classifying galaxies into three environmental categories: main cluster, group, and neither, using observable galaxy properties such as positions, line-of-sight velocities, and stellar mass. The model is trained on mock observations derived from cosmological simulations designed to replicate survey conditions and achieves an overall accuracy and class-size-weighted precision of 81%. The neither and main cluster populations are reliably recovered, whereas group galaxies remain the most challenging to identify, achieving 30% completeness and 76% purity. Within $1\times R_{200}$, classification performance is suppressed, but it improves beyond this radius, reaching 40% completeness and 80% purity. Resampling and thresholding strategies allow the model to be tuned toward either higher purity or higher completeness; in this study, we adopt first-past-the-post thresholding to emphasise purity. Model performance is consistent across cluster masses and dynamical states, and it outperforms both Friends-of-Friends and Gaussian Mixture Modelling. This flexibility makes it well suited to upcoming spectroscopic surveys of cluster infall regions, providing a robust statistical tool for disentangling environmental influences on galaxy evolution.