This manuscript introduces ML-proxkNN, a spatially-aware extension of multi-label kNN that integrates (i) locally adaptive neighbourhood size and (ii) spatial proximity/context via Moore neighbourhood information. Results on Milan and Trento show strong multi-label performance (up to 0.920 average precision on the combined Milan data source) and improved spatial coherence, with gains that are often incremental but consistent across metrics and validated under spatial block cross-validation.