Tile manufacturers and facility managers verify slip resistance using walkway tribometers, which require device setup, surface conditioning, and repeated strut releases for each measurement. Surface roughness can be captured in seconds with a portable profilometer. This study evaluates whether profilometry, combined with gradient boosting, can replace a share of routine friction testing. Three gradient boosting algorithms, Gradient Boosting (GB), XGBoost, and LightGBM, predict the coefficient of friction (COF) of ceramic floor tiles from six roughness parameters and shoe hardness, with Linear Regression and Random Forest as baselines. Measurements were taken on seven ceramic tiles, four shoe materials, and three surface conditions (dry, wet, and soapy), yielding 2,016 observations. Models were evaluated under two cross-validation protocols: random five-fold cross-validation, which quantifies interpolation within characterized surfaces, and leave-one-tile-out grouped cross-validation, which quantifies generalization to unseen tiles. Hyperparameters were tuned with a nested grid search, and algorithm differences were tested using paired t-tests on fold scores. Under the interpolation protocol, GB and XGBoost were practically equivalent (R² up to 0.971 ± 0.014 in wet conditions, mean differences below 0.004) and outperformed LightGBM and both baselines; the linear baseline reached R² of only 0.21 to 0.39. Under the generalization protocol, no model on this set of seven tiles extrapolated to a held-out tile (pooled R² below zero), a finding that highlights the need for grouped validation when claiming model transferability and delimits roughness-based screening to tile families represented in the training data. SHAP analysis identified maximum height (R_z) and shoe hardness as the dominant predictors. On laboratory-measured samples, the approach shows promise for quality control of characterized tile families; uncalibrated transfer to new surfaces is cautioned against.

Keywords

Slip resistance, coefficient of friction, gradient boosting, machine learning, surface roughness.