Micro-textured water-lubricated thrust bearings exhibit significant potential in motorized spindles due to their low friction, high heat dissipation efficiency, and superior damping performance. However, existing numerical methods for evaluating the lubrication performance of such bearings face challenges in balancing computational efficiency with accuracy, along with limitations in macro-micro cross-scale coupling capabilities. To address these issues, this study proposes a novel approach combining the flow factor model with a machine learning algorithm. First, the average Reynolds equation based on the average flow model (AF-ARE) is formulated using flow factors. However, results indicate that when dealing with high-density textures with small diameters, AF-ARE still suffers from computational inefficiency as it requires individual calculation of flow factors for each texture element. To address this limitation, machine learning-based prediction models are subsequently developed using three algorithms: Gaussian process regression, support vector machine, and extreme learning machine. The prediction models enable rapid estimation of flow factors for all texture elements, and their predictive performances are systematically compared and evaluated. Subsequently, by integrating these flow factor prediction models with the average Reynolds equation, a novel average flow-machine learning-averaged Reynolds equation (AFML-ARE) method is proposed. The proposed AFML-ARE multiscale numerical method offers a novel approach to overcome current research limitations in cross-scale numerical analysis of lubrication characteristics for high-density, small-diameter micro-textured bearings, enabling efficient performance evaluation from microscopic texture effects to macroscopic bearing behavior.