<p>Artificial ventilation supports patients needing respiratory assistance, where an effective and reliable control mechanism is vital due to varying patient conditions, lung compliance, and resistance. Traditional controllers like proportional–integral–derivative often encounter challenges such as overshoot, long settling times, noise sensitivity, and instability at high frequencies. This study addresses these limitations by proposing a novel and robust neural-based control solution that adapts to diverse clinical conditions, which is crucial for ensuring patient safety and ventilator efficiency. To address these issues, a shallow neural network-based controller (SNNC) is proposed for ventilator pressure control. Training neural networks typically involves stochastic gradient descent with backpropagation, which may converge to sub-optimal solutions and requires differentiable activation and loss functions. To overcome these limitations, a wavelet mutated class topper optimization algorithm (WmCTO) is developed to train the SNNC, termed WmCTO-SNNC. The wavelet mutation operation helps avoid local optima inherent in the classical CTO. The proposed WmCTO-SNNC controller demonstrated superior performance in extensive simulations of a patient-hose ventilator model. Key improvements include lower overshoot (19.987 mbar vs. 23.87 mbar in baseline), faster settling time (0.08s vs. 0.25s), and smoother control signal behavior. The WmCTO-SNNC effectively tracks ventilator pressure and conserves energy by reducing control signal fluctuations and oscillations. Furthermore, it proved robust across virtual patient models with varying lung characteristics and responded efficiently to pressure disturbances, confirming its adaptability and real-time applicability.</p>