Relative performance of multiple machine learning algorithms for land cover image classification in an urbanized landscape over Southwest Ethiopia: implications for landscape management

Sustainable economic and environmental stability in Ethiopia’s rapidly urbanizing regions requires an accurate understanding of land use and land cover (LULC) changes, which necessitates the deployment of effective classifier algorithms. This study assessed the performance of four image classifier algorithms: Random Forest (RF), Support Vector Machine (SVM), Classification and Regression Tree (CART), and K-Nearest Neighbor (KNN) in producing LULC maps and quantifying urban expansion intensity in Jimma City, southwestern Ethiopia, from 1993 to 2023. Maps for 1993, 2008, and 2023 were classified using Google Earth Engine and Landsat satellite imagery. The classifiers exhibited distinct performances, with SVM achieving the highest accuracy (Overall Accuracy (OA): 0.87 to 0.93; Kappa: 0.89 to 0.91), followed by RF (OA: 0.82 to 0.87; Kappa: 0.79 to 0.84), KNN (OA: 0.78 to 0.81; Kappa: 0.74 to 0.81), and CART (OA: 0.77 to 0.83; Kappa: 0.79 to 0.81). The LULC analysis using SVM revealed significant changes: built-up areas increased (2.1% to 23.1%), grass/shrubland varied (36.55% to 33.15%), and agricultural land declined (37.1% to 18.0%). Other classes, including water bodies (0.22% to 0.42%), forestland (23.83 to 24.16%), and barren land (0.02% to 0.76%), also exhibited fluctuations. The urban expansion intensity indices for 1993–2008 (0.787), 2008–2023 (0.607), and 1993–2023 (0.590) indicate rapid urban expansion. The study concluded that Jimma City is experiencing medium-speed urban expansion, with implications for ecosystem stability and resource management. To promote sustainable urban growth, the city should implement further research programs and adopt strategies such as compact development, green space preservation, and integrated land-use zoning. Rapid urbanization in Jimma City, Ethiopia, necessitates accurate land cover mapping for sustainable management. Four major machine-learning classifiers (SVM, RF, CART, and KNN) were evaluated on Google Earth Engine using Landsat imagery. The Support Vector Machine (SVM) algorithm outperformed others, achieving the highest accuracy (up to 93% Overall Accuracy). Significant land use dynamics were observed, with built-up area expanding from 2.2% (1993) to 23.1% (2023), indicating intense urban growth.