Visual field (VF) testing is an important diagnostic tool for ophthalmologists to assess the extent of vision loss and the affected regions. VF loss often follows distinctive spatial patterns that are indicative of specific ocular and neurological conditions. Accurate identification and characterization of these patterns is critical for disease classification and progression monitoring. However, existing VF datasets typically lack annotated labels for these patterns, and manual labelling is resource-intensive. Moreover, there remains a lack of research focused on benchmarking generalizable archetypes across different datasets. Therefore, this research proposes a methodology grounded in an unsupervised learning approach, specifically Archetypal Analysis, to identify representative patterns of visual field loss. The proposed method is evaluated using two datasets: UWHVF and eFOVID. The resulting generalized VF patterns are found to be consistent across both datasets. Furthermore, the identified patterns are reviewed by ophthalmologists and found them as clinically relevant.