Direct urea fuel cell (DUFC) has recently gained attention as a potentially effective fuel cell because of urea’s ease of use, low-cost, high-energy density, and toxicity-free nature. To develop a fast and accurate mathematical tool to model and optimize the performance of DUFC, this study, for the first time, combined the kinetic model and artificial intelligence approaches. Accordingly, an artificial neural network (ANN) model was developed based on the synthetic datasets produced from the verified kinetic model and then was optimized by particle swarm optimization to determine the optimal conditions for DUFC. The peak power density output and substrate-based energy recovery were chosen as the target functions that depend on several fundamental operating and designing parameters (i.e., types of catholyte, membrane thickness, diffusion layer thickness and porosity, feeding flow rate, temperature, urea and KOH concentrations). Along with significantly reducing the computational simulation time to the kinetic model (from 1500 s to 2 s), the proposed approach exhibited superior accuracy compared to the response surface methodology. With nickel nanoparticles as the anode catalyst material and pure oxygen as the catholyte, the highest power density output and energy recovery from DUFC were predicted at 15.1 mW/cm2 and 19.6 %, respectively. On the other hand, with the natural air supply, the optimal power density and energy recovery values were only 14.4 mW/cm2 and 18.9 %, respectively, due to lower oxygen concentration. These poor performances suggest that more efficient catalyst materials for DUFC need to be further investigated and applied.