<p>Bio-trickling filter (BTF) hasemerged as a sustainable and cost-effective solution for H₂S removal. However, the performance of BTF is sensitive to operating parameters and challenging to optimize due to the nonlinear and dynamic behaviour of biological systems. In this study, a novel hybrid modelling approach was developed using a multi-input-multi-output (MIMO) framework combining Box-Behnken Design (BBD), back-propagation artificial neural network (BP-ANN), and a genetic algorithm (GA) to optimize H₂S removal efficiency (<i>RE</i>) and elimination capacity (<i>EC</i>) in a lab-scale BTF unit. Key process variables such as H₂S concentration, empty bed residence time (EBRT), and liquid trickling rate (LTR) were systematically studied. The BP-ANN model was trained on experimental data generated via BBD, and its optimized weights were used as the initial population for GA. Model performance was evaluated using sum of square error (<i>SSE</i>), mean square error (<i>MSE</i>), coefficient of determination (<i>R</i>^<i>2</i>), relative percent error (<i>RPD</i>), and root mean square deviation (<i>RMSE</i>) metrics. The optimized conditions achieved with maximum <i>RE</i> and <i>EC</i> with less than ± 1.5% error compared to experimental results. This integrated modelling and optimization approach demonstrated high predictive accuracy and operational reliability, which offering a robust tool for process intensification of gas-phase biofiltration systems.</p>