Comparative study of AI algorithms in boundary layer flow: Evaluating performance of levenberg-marquardt, bayesian, and scaled conjugate methods

Abstract

The current research article presents an investigation into steady two-dimensional incompressible Sisko fluid flow, which is produced by magnetohydrodynamics (MHD), thermal radiation, and bioconvection over an exponential sheet. The obtained partial differential equations are reduced to coupled ordinary differential equations using appropriate similarity variables. The numerical results are first achieved through the BVP4C technique. After that, an artificial neural network will be performed to test the ability and accuracy of the model using the dataset in training, validation, and testing sets. The study mainly focuses on testing artificial neural networking (ANN) algorithms such as Levenberg-Marquardt, Bayesian regularization, and scaled conjugate gradient. After concluding the previous steps, it is found that the most suitable algorithm for this project is the Levenberg-Marquardt algorithm as it is efficient. Table 4 show the statistical difference between these three AI techniques. The research proves that an increase in fluid parameters improves the velocity profile, whereas greater thermal radiation strengthens the energy profile. This indicates the thermal and flow characteristics of Sisko fluid, where momentum and heat transfer are improved with these parameters. Both the Bayesian and Scaled Conjugate algorithms exhibit low performance; Scaled Conjugate displays fluctuating fitness, while Bayesian lacks validation data. Levenberg-Marquardt ensures accurate modeling by achieving high regression (∼1) and ideal MSE (≤10−8). Sisko fluid on a sheet containing gyrotactic bacteria improves bioconvective heat transmission and has applications in microfluidics and biomedicine.