Orthogonal experimental design and random forest machine learning were integrated to optimize the low-friction performance of WC/a-C films at 15% RH. Range analysis determined the optimal parameters: 5.43 at. % W content, 5   N load, 6   Hz sliding frequency, and GCr15 counterpart ball, achieving a friction coefficient of 0.026. Microstructural investigations revealed that parameters modulated friction through influencing C/WO 3 sliding interface formation. The Random Forest model corroborated the orthogonal experimental design-derived importance hierarchy (load > frequency > W content). Predictions using this model yielded a friction coefficient of 0.032 for the optimal parameters, significantly reducing the experimental effort required for optimization. This methodology effectively advanced low-friction design of hydrogen-free carbon-based films under low-humidity atmospheric conditions.