<p>This study integrated Object-Based Image Analysis (OBIA) with machine learning algorithms within the ‘tidymodels’ framework in R to enhance land use and land cover (LULC) classification using PlanetScope imagery. Despite the increasing adoption of OBIA and machine learning in LULC mapping, few studies have systematically compared multiple algorithms under a unified and statistically rigorous framework. Six classifiers, including Random Forest, Extreme Gradient Boosting, Artificial Neural Network, Support Vector Machine, K-Nearest Neighbors, and Naïve Bayes, were used to classify five land cover types: greenhouse, bare land, built-up land, vegetation, and water bodies. OBIA extracted 12 spectral and geometric predictors in ArcGIS, while ‘tidymodels’ facilitated data pre-processing, resampling, hyperparameter tuning, and model assessment. Model performance was evaluated using 50 bootstrap samples focusing on the distributions of hard prediction metrics (overall accuracy, precision, recall, F1-score, and Kappa) as well as soft prediction metrics (ROC-AUC and mean log loss). XGBoost achieved the best performance (F1-score = 0.84 ± 0.0034, Kappa = 0.84 ± 0.0023, ROC-AUC = 0.98 ± 0.0006, mean log loss = 0.33 ± 0.0048), indicating high accuracy and confidence in classification. Random Forest ranked second, followed by Artificial Neural Network. The two models showed higher mean log loss despite strong accuracy. The ‘yardstick’ package in tidymodels enabled robust statistical comparisons, demonstrating the stability and confidence of ensemble-based classifiers. This study proposes a reproducible framework that aggregates OBIA and ‘tidymodels’ for reliable LULC mapping, supporting sustainable urban and agricultural land management.</p>