Steam reforming of hydrocarbons is and will be, in the short-medium run, the leading technology for producing hydrogen. At the same time, steam reforming has a large carbon footprint that can be decreased by implementing better kinetic models for process intensification. In this work, we have developed a methodology based on artificial neural networks to fit and improve the robustness of the kinetic model for steam reforming of naphtha surrogates (hexane and heptane) using a NiMgAl catalyst derived from hydrotalcite precursors. We analyze several strategies to obtain the fittest kinetic model and discuss the robustness of each. These models include hydrocarbon steam reforming, water gas shift and methanation reactions, and differ mainly in the type of adsorption term in the Langmuir-Hinshelwood formalism. The adsorption energies calculated by ab initio (DFT) provide insights on the different adsorption mechanisms of hydrocarbons and water on the catalyst surface sites. At the same time, validation of the kinetic model was conducted using wider range of experimental conditions and different model and real feeds (methane, naphtha, diesel and vegetable oil). In this way, the versatility of the model proposed and the strengths and weaknesses of using a data-driven approach for the kinetic model selection were proven.