Due to the precision of its results, the use of artificial intelligence technology for predictive models for multiphase flow description currently has a considerable impact on industrial applications. This work compares the results obtained in an adaptive neuro-fuzzy inference system with the results obtained in an artificial neural network to predict the holdup of oil in vertical pipes with a water and oil mixture. We use pipe diameter, fluid surface velocity, and oil viscosity as input parameters. The research was conducted using 722 experimental data and found that the best model is a model generated by an artificial neural network with a hidden layer consisting of 12 neurons and using the Levenberg Marquardt training function and the TanSig activation function. The results showed a root mean square error of 0.000049, the coefficient R2 was 0.99932, and the average absolute percentage error was 0.19%. The predictive model can be used to improve transportation processes in the oil and gas sector.