While the ability to use ceramic waste as a concrete aggregate and cement substitute provides a sustainable path to more carbon neutral concrete, the non-linear relationship between the performance of a concrete made from ceramic waste and the ceramic waste content makes the design of such concrete more complicated. This study proposes and tests a computational (machine-learning) tool to forecast the compressive strength of ceramic waste-based concrete based on mix-proportion and curing parameters, and then compares the performance of the model. A structured data set was compiled and pre-processed with a total of 8 input features: binder content, ceramic waste powder (CWP) replacement, water-binder ratio, fine and coarse aggregate, ceramic coarse-aggregate replacement, superplasticiser dosage, curing age. Five predictive models were trained and compared, multiple linear regression (MLR), support-vector regression (SVR), an artificial neural network (ANN), random forest (RF) and extreme gradient boosting (XGBoost). The evaluation of models was done with an independent test set by R², RMSE, MAE and MAPE. The accuracy of XGBoost was also the best (test R² = 0.944, RMSE = 2.73 MPa, MAPE = 5.0%) compared to the baselines of the ANN, RF, SVR and MLR. Feature-importance and sensitivity analyses revealed that the water-binder ratio and the curing age are the primary drivers, whereas the replacement with CWP showed a clear optimum around 18%, with the strength predicted decreasing above this level, which can be explained by the pozzolanic-then-dilution behaviour of this variable experimentally reported. The validated model is able to reproduce experimental trends within ±10%, and serves as a fast and inexpensive screening tool for ceramic-waste mixtures and to assist in sustainable mix design, avoiding the need for comprehensive laboratory testing.