ANN-assisted analysis of Hall-MHD Casson blood–gold nanofluid flow over a stretching sheet

This study develops a validated numerical-neural framework for the steady boundary-layer flow of a Casson blood-gold (Au) nanofluid over a linearly stretching sheet. The model includes a transverse magnetic field with Hall current, Rosseland thermal radiation, Brownian motion, thermophoresis, internal heat generation and an Arrhenius-type chemical reaction. The mixture properties are computed with the Tiwari-Das model, while Buongiorno's slip mechanisms govern the local nanoparticle concentration. Similarity variables reduce the equations to a coupled nonlinear boundary-value problem, which is solved by fourth-order Lobatto IIIA collocation with residual control (the bvp4c algorithm). The solver reproduces the benchmark data of Khan and Pop to four decimal places and an exact Casson-MHD-nanofluid solution to within 10⁻⁶ . The numerical solutions (17 710 samples in ten scenarios) are used to train single-hidden-layer (2-12-1) networks with the Levenberg-Marquardt (LM) algorithm. The networks reach test mean-squared errors of 10⁻¹¹ - 10⁻⁸ , regression coefficients R > 0.99999 and maximum absolute errors below 1.1× 10⁻³ at unseen parameter values; for the same network, LM gives errors two to five orders of magnitude lower than Adam or stochastic gradient descent. Adding 4% gold to blood raises the heat-transfer rate by 8.0% and lowers the streamwise wall shear by 17.9%, because the gold loading strengthens the thermal buoyancy. Increasing the magnetic parameter from 0.5 to 3 raises the streamwise wall shear by 86.5%, thermal radiation ( R : 0.1→1.5) doubles the wall heat flux (+112%), and the chemical reaction ( K : 0.5→2) raises the Sherwood number by 133%. The work is a fundamental transport model and a fast surrogate for the parametric screening of magnetically controlled blood-nanoparticle heat transfer.