This study focuses on the prediction of hydrogen sulphide removal efficiency in natural gas sweetening processes using Artificial Neural Networks (ANN). The developed model was trained on experimental solubility data that were extracted from literature for 17 different absorbents at various operating conditions, with a total of 470 data points. Three training algorithms of Scaled Conjugate Gradient (SCG), Levenberg Marquardt (LM), and Bayesian Regularization (BR) were evaluated to determine the optimal ANN architecture. The performance of each model was assessed using mean squared error (MSE) and the coefficient of determination (R²). The SCG algorithm achieved its best performance at 35 hidden neurons, with MSE = 0.0183 and R² = 0.8799, showing gradual improvements but lower accuracy compared to other methods. The LM algorithm performed optimally at 15 hidden neurons, yielding MSE = 0.002865 and R² = 0.9785, demonstrating excellent predictive accuracy. The BR algorithm outperformed both SCG and LM, with the best results at 25 hidden neurons, achieving MSE = 0.001465 and R² = 0.9904, indicating superior generalization and stability. These results highlight the potential of ANN as a robust tool for simulating gas sweetening processes and supporting industrial decision-making.