The application of machine learning has become very relevant in the field of heat transfer as it helps in the prediction and optimization of heat flow analysis in heat transfer devices. This study uses machine learning approach to identify the most accurate optimal assumption for heat transfer coefficient in various typical nanofluid configurations. In the study, algorithms were developed using information from nanofluid experiments and models in literature. MATLAB was used as the machine learning language to create a technique for forecasting the percentage error in heat transfer coefficient of a specific nanofluid model with experimental dataset as input. An optimization model was developed from the algorithm to reduce simulation error and return the simulation's best possible assumptions for different configurations. The root mean squared error was the performance parameter employed to assess the algorithm's correctness. The single-phase, discrete phase, Eulerian, mixture, combination model of discrete and mixture phases, the volume of fluid, dispersion, and Buongiorno's model were the eight models examined. The results revealed that the optimization model accurately identified the best assumptions for different nanofluids configurations. This is an important decision-making approach in choosing the assumptions best suited for different nanofluid configurations simulations.