Bayesian-regularized neural network modeling of ternary hybrid Ellis nanofluid flow over a radiative surface

Purpose This study aims to analyze the mixed convection flow behavior of Ellis nanofluid over a stretching cylinder using an Artificial Neural Network (ANN) based on the Levenberg–Marquardt scheme. The objective is to investigate how different physical parameters influence velocity, temperature, skin friction and heat transfer characteristics of the nanofluid system under varying nanoparticle concentrations. Design/methodology/approach The governing physical model is transformed into nonlinear ordinary differential equations using similarity transformations. These equations are solved numerically with the MATLAB solver BVP4C to generate a reference data set. The data set is then used to train an ANN with the Levenberg–Marquardt backpropagation algorithm. Data are divided into 70% training, 15% validation and 15% testing to ensure reliable model evaluation. Findings The results of this study indicate that the mean squared error decreases steadily with increasing training epochs, confirming effective learning of the ANN model. Error histograms show very low mean errors, demonstrating high prediction accuracy. It is also observed that the velocity profile increases with the nanoparticle material parameter, while the temperature distribution rises with higher heat generation parameter values. Originality/value This study presents a hybrid computational approach that integrates numerical solutions with ANN modeling to analyze Ellis nanofluid flow over a stretching cylinder. The combination of the Levenberg–Marquardt ANN with numerical benchmark data provides an efficient framework for predicting thermal and flow characteristics, offering an alternative data-driven methodology for complex nanofluid flow problems.