Artificial neural networks have reshaped machine learning by delivering unparalleled proficiency in unraveling intricate phenomena and addressing multifaceted challenges. Backpropagation remains the foundation for training these networks, but its optimization is essential for tackling sophisticated fluid dynamics problems. This study leverages the Levenberg–Marquardt technique integrated with artificial neural network backpropagation (LMT-BP-ANN) to examine the behavior of radially magnetized boundary layers in a unique nanofluid. Composed of nickel \(\left(Ni\right)\) , molybdenum disulfide \(\left(Mo{S}_{2}\right)\) and tantalum \(\left(Ta\right)\) nanoparticles this nanofluid is investigated as it flows across a curved surface while managing heat dissipation and frictional forces with ethylene glycol serving as the base fluid. The fluid model is optimized by methodically varying critical physical parameters including the magnetic parameter, Reiner–Philippoff parameter, Bingham number, curvature parameter, Eckert number and thermal radiation parameter. The local non-similarity method implemented in MATLAB’s bvp4c solver generates the dataset for the LMT-BP-ANN framework. The neural network employs 70% of the data for training and 15% for validation. Performance plots, regression analyses and error histograms illustrate the precision, efficiency and robustness of the proposed LMT-BP-ANN architecture. The network achieves a mean squared error on the order of \({10}^{-10}\) and an R-squared value of 1 during operation. The analysis further reveals that the fluid’s velocity diminishes as the magnetic parameter increases.
