Synergistic comparison of ANN and ANFIS predictive modeling with experimental electrochemical performance data in CO₂ conversion

This work systematically evaluated CO₂ conversion performance metrics across the recent high-impact literature by focusing on five key inputs; applied potential (AP), catalyst loading (CL), composition (CO), support (SU) and electrolyte (EL) and two outputs; Faradaic efficiency (FE) and current density (CD). Dataset Sixty-four series were evaluated to generate predictive models for FE and CD. Residual plots showed a good fit along the fitted axis with deviations from normality being quite small, confirming linear relationships between input and output. Histograms have shown most of the errors were on the ranges - 0.8 to 0.8 for FE and - 4 to 4 for CD, thus confirming that the error was normally distributed and improving prediction reliability. Optimal electrode with reference to SN ratio was determined to be AP2-CL1-CO1-SU 2-E12 (-1.29 V, 0.426 mg/cm², 10% of Cu-In, carbon papers support and 0.1 M KOH) as optimal for increasing Faradic efficiency and AP1-CL1-CO1-SU1-EL1 (-1.29 V, 0.426 mg/cm², 10% of Cu-In, graphene papers and 0.1 M KHCO3) as optimal for improving Current Density. ANOVA results showed that the catalyst loading and applied potential are two major factors influencing FE and CD.ANN (5-10-2) exhibited good prediction performance (overall R² = 0.93726). From the performance metrics, ANFIS exhibited low error values RMSE (0.0725 & 0.0525) and MAE (0.1411 & 0.0811) for both FE and CD when compared with ANN and also the corresponding ΔR² confirmed that ANFIS delivered the better predictive accuracy. This comparison demonstrated the strong predictive power of the ANFIS with respect to ANN, verifying the reliability of the modelling approach based on experimental results.