Knee osteoarthritis is a widespread and debilitating joint disorder affecting millions of individuals globally, leading to chronic pain, stiffness, and reduced mobility. Traditional diagnostic methods, although effective, can be resource-intensive and not always accessible. This study aims to develop and evaluate a markerless machine learning approach for classifying gait patterns in individuals with KOA and healthy controls. Utilizing the VitPose model for pose estimation, we extracted spatiotemporal and kinematic features from gait videos and employed machine learning classifiers to achieve high classification accuracy. Among the classifiers tested, the Random Forest model demonstrated superior performance with an accuracy of 0.944, sensitivity of 0.904, specificity of 0.965, and a Matthews Correlation Coefficient of 0.875. These results outperform previous studies using the same dataset, highlighting the effectiveness of our approach. This markerless technique offers a more accessible and efficient tool for early and accurate diagnosis of KOA, ultimately improving patient outcomes. Future research should focus on expanding the dataset and validating the models in diverse clinical settings to enhance the generalizability of the findings.