The identification of producer typologies is a crucial tool for understanding the heterogeneity of agroecosystems and designing targeted policies. Andean agroecosystems, particularly those in rapidly suburbanizing areas, have been understudied in this regard, creating a critical knowledge gap. This study addressed this void by determining the typologies of smallholder agroecosystems in the suburban periphery of Cuenca, Ecuador, by applying an unsupervised machine learning technique, Partitioning Around Medoids (PAM) Clustering, to survey data from 293 farmers. Our analysis revealed three distinct typologies, highlighting a socio-economic and productive gradient defined by income sources, market access, and agrochemical use. The typologies range from economically vulnerable households to more commercially oriented and environmentally sustainable ones, underscoring the complex interplay between livelihoods strategies and environmental management. This research provides one of the first empirical typologies of suburban Andean agroecosystems, demonstrating the value of unsupervised learning for capturing farm heterogeneity in data-scarce contexts. The findings offer a robust evidence base for moving beyond one-size-fits-all approaches, enabling the design of differentiated agricultural and territorial policies that enhance sustainability, equity, and resilience at the rural–urban interface.