Performance comparison of machine learning algorithms for land cover classification

Accurate land use and land cover (LULC) classification is essential for environmental monitoring, resource management, and policy planning. This study evaluates the seasonal performance of six machine learning (ML) algorithms, including random forest (RF), support vector machine (SVM), k-nearest neighbor (k-NN), gradient tree boost (GTB), classification and regression tree (CART), and naive Bayes (NB), using the Google Earth Engine (GEE) platform. The analysis was conducted in Butler County, Ohio, a temperate region with distinct seasonal variability, using optical imagery across spring, summer, fall, and winter. Classifier performance was assessed using overall accuracy (OA), kappa coefficient, and F1-score. Results indicate that classification performance exhibits seasonal variation associated with phenological dynamics, with spring showing the highest average OA (93.84%), kappa (92.30%), and F1-score (0.94), highlighting how seasonal surface conditions influence relative algorithm performance. Among the ML classifiers, k-NN achieved the highest average OA (95.76%), kappa (94.7%), and F1-score (0.96), closely followed by RF with OA (95.45%), kappa (94.32%), and F1-score (0.95), while NB exhibited substantially lower performance across all seasons (average OA: 78.94%, kappa: 73.68%, F1-score: 0.78). These findings underscore how seasonal surface conditions influence relative algorithm performance and highlight the importance of season-aware model selection for reliable year-round LULC classification. The results further demonstrate how seasonal analysis can inform data acquisition timing and support the design of land-monitoring programs across diverse environmental contexts.