Objective: This study aimed to apply machine learning approaches to the Kellgren-Lawrence (KL) grading of knee osteoarthritis, develop an effective automatic KL grading technique, and provide a methodological reference for clinical diagnosis and research. Methods: Data were obtained from the Osteoarthritis Initiative (OAI) knee X-ray image dataset, comprising 8,110 images from the folders of auto_test, train, and val. All images were first subjected to inversion processing, followed by extraction of two-dimensional radiomic features. Feature selection was then conducted using a combination of variance thresholding and analysis of variance (ANOVA), yielding 18 key features. To address class imbalance in the original dataset, this synthetic minority over-sampling technique (SMOTE)  and class weight balancing were jointly applied. Eight machine learning models-Decision Trees (DT), Logistic Regression (LR), Random Forests (RF), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), and Light Gradient Boosting Machine (LightGBM)-were trained for KL grading of knee osteoarthritis. Model performance was evaluated using accuracy, precision, recall, F1-score, and the area under the curve (AUC). For the optimal SVM model, global and local interpretability analy-ses were further conducted using SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) to identify the key factors influencing model decisions. Results: The support vector machine  (SVM) model achieves the best performance. Conclusion: This study establishes an effective machine learningbased method for automatic KL grading of knee osteoarthritis, providing valuable support for clinical diagnosis and research applications.