The removal of cyclophosphamide and ibuprofen from aqueous solutions was investigated through adsorption-photocatalytic experiments combined with kinetic, isotherm, thermodynamic, and artificial neural network modeling. Nonlinear kinetic analysis showed that the pseudo-second-order model was more reliable for cyclophosphamide (R2 = 0.9954), while the pseudo-first-order model showed better agreement for ibuprofen (R2 = 0.9898) with 95% confidence intervals for the estimated parameters. Equilibrium results indicated that cyclophosphamide removal was well described by the Temkin model (R2 = 0.9770), with the Langmuir model also showing good agreement (R2 = 0.9740), whereas ibuprofen removal was better represented by the Freundlich isotherm (R2 = 0.9960), suggesting heterogeneous surface interactions. Thermodynamic parameters revealed spontaneous and endothermic overall removal processes, characterized by negative ΔG° and positive ΔH° (138.37 kJ mol−1 for cyclophosphamide and 83.76 kJ mol−1 for ibuprofen) with confidence intervals indicating reliable estimation. A multilayer perceptron artificial neural network was developed to predict removal efficiency using operational variables as inputs. The optimized networks contained 14 and 18 hidden neurons for cyclophosphamide and ibuprofen, respectively. The network demonstrated excellent predictive performance, with overall correlation coefficients of 0.99839 and 0.99587 and low prediction errors. Robustness was confirmed through repeated 5-fold cross-validation, yielding mean R values of 0.99141 ± 0.00669 for cyclophosphamide and 0.97944 ± 0.02108. Comparison with support vector regression demonstrated the superior predictive capability of the ANN on the test datasets. The strong agreement between predicted and experimental results highlights the reliability of the proposed framework for modeling pharmaceutical removal under irradiated conditions and optimizing treatment processes for emerging contaminants.
