The performance of a flue gas desulfurization (FGD) system is characterized by SO₂ removal efficiency ( $Y_1$) and reagent conversion ( $Y_2$). Achieving a near-perfect reaction environment has been of concern in dry FGD (DFGD) due to the low reactivity compared to the wet and semi-dry units. This study will appraise output responses using modeling by response surface methodology (RSM) and artificial neural networks (ANN) approaches. The impacts of input parameters like hydration time, hydration temperature, diatomite to hydrated lime (Ca(OH)₂), sulfation temperature and inlet gas concentration will be studied using a randomized central composite design (CCD). ANN fitting tool mapped the CCD metadata using the Levenberg-Marquardt (LM) algorithm activated by the hyperbolic tangent (tansig) function. The hidden cells ranged from 7 to 10 to ascertain the effect node architecture on modeling accuracy. Validation of each procedure was assessed using root mean square error (RMSE), mean square error (MSE) and R-Squared studies. The outcomes presented a more accurate 5-10-2 ANN model in the mapping of the DFGD from R² data of $Y_1$ = 0.993 and $Y_2$ = 0.9986 with a mapping deviation from the RMSE values of $Y_1$ = 0.48465; $Y_2$ = 0.44971 and MSE results of $Y_1$ = 0.23488; $Y_2.$= 0.20229.

Perceptron models have become integral tools for pattern recognition and classification problems in engineering fields. This study envisioned implementing artificial neural networks to forecast the performance of a mini-spray dryer for desulfurization activities. This work adopted k-fold cross-validation, a rigorous technique that evaluates model performance across multiple data segments. Several ANN models (21) were trained on data obtained from sulfation conditions, including sulfation temperature (120 °C-200 °C), slurry pH (4-12), stoichiometric ratio (0.5-2.5), slurry solid concentration (6%-14%) as the feed input and sulfur capture as the response. Three hundred synthetic datasets generated using the Gaussian noise data augmentation underwent a 10-fold cross-validation process before simulation on neural networks triggered by the logsig and tansig activation functions. The computation accuracy was further evaluated by altering the number of hidden cells from 2 to 10. The ANN architectures were assessed using statistical metrics such as mean square error (MSE), root mean square error (RMSE), mean absolute percentage error (MAPE), and the coefficient of determination (R²) techniques. Overall, error estimation suggests cross-validation and data augmentation are critical in efficient neural network generalization. The logsig function trained with 10 hidden cells presented closer data articulation when mapped onto actual values.