Sonocatalysis, an efficient technique for the degradation of organic contaminants, synergistically combines ultrasonic irradiation with catalysts such as CeO2 nanorods (NRs). This study investigates the utilization of CeO2 NRs in this process to degrade caffeine in water. The CeO2 NRs was synthesized via the hydrothermal method and characterized using XRD, high-resolution transmission electron microscope (HRTEM), Fourier-transform infrared spectroscopy (FTIR), and Raman spectroscopies. Then, the effect of the process parameter on the sonocatalytic degradation of caffeine was evaluated by varying the initial pH of the solution, initial caffeine concentrations, and catalyst dosages. Machine learning (ML) models, including support vector machines (SVM), artificial neural networks (ANN), and Gaussian process regression (GPR) with bootstrap resampling, were employed to analyze and optimize the process. The bootstrap technique improved prediction accuracy by generating additional data, with the GPR-bootstrap model achieving an R2 of 0.999 and RMSE of 0.001. The highest caffeine degradation (91.5%) was observed at an initial pH of the solution of 7.5, 1 g L−1 of CeO2 NRs dosage, 5 mg L−1 of caffeine concentration, and 20 mM H2O2. These findings confirm the effectiveness of bootstrap-enhanced machine learning in accurately predicting and optimizing the sonocatalytic processes.