Comparative analysis of intelligent predictors using multi-supervised learning models for bioconvective flow of Oldroyd-B nanofluid over a Riga surface in a porous medium

Efficient control of heat and mass transfer in complex, non-Newtonian fluids is central to improving the energy efficiency of modern thermal management systems, yet such flows remain difficult to model because they couple fluid motion, heat transport, and the swimming behavior of motile microorganisms. Our model describes the thermal bioconvective flow of a non-Newtonian nanofluid containing gyrotactic microorganisms over a porous stretching Riga surface and asks whether artificial intelligence-based predictors can reproduce this behavior more quickly and reliably than purely numerical solutions. The governing equations were first reduced to a system of ordinary differential equations through similarity transformations, then solved with a fourth-order collocation-based boundary value solver. The resulting data set was used to train and compare four supervised machine learning predictors: a long short-term memory network, a nonlinear autoregressive network with exogenous inputs, a random forest regressor, and a support vector regressor. Across all scenarios, the random forest predictor performed best on every metric, achieving the highest coefficient of determination (average 0.9876, reaching 0.9965) and the lowest error, with a mean squared error of 6.28 × 10 −7 and a root mean squared error of 0.0007. It was followed by the long short-term memory network (coefficient of determination 0.9508), the nonlinear autoregressive network (0.9412), and the support vector regressor (0.9286). The close agreement between our machine learning predictions and the numerical solutions confirms that the proposed framework is accurate, stable, and computationally efficient. The novelty of this work lies in combining a detailed numerical simulation of a bioconvective non-Newtonian nanofluid with a systematic comparison of several supervised learning models, an approach that goes beyond earlier studies relying on a single method or purely numerical analysis, and offers a practical route toward intelligent modelling of industrial cooling, electronic thermal control, and biomedical fluid systems.