A novel predictive deep learning network design for dynamics of doubly stratified micropolar-Casson fluid flow model

Purpose This study aims to investigate the behavior of doubly stratified micropolar-Casson fluid (CF) flow over a stretching sheet in a porous medium, focusing on heat and mass transfer characteristics in non-Newtonian systems. This work uses deep autoregressive exogenous neural networks optimized via the Levenberg–Marquardt method (DARX-NNs-LMT) to model and predict the underlying nonlinear dynamics. By transforming governing partial differential equations into ordinary differential equations and analyzing the effects of key physical parameters, this study seeks to provide an accurate and efficient computational framework for understanding complex fluid behavior in biomedical and industrial applications. Design/methodology/approach The governing partial differential equations describing the doubly stratified micropolar-CF flow are transformed into a nonlinear system of ordinary differential equations using appropriate similarity transformations. A data set is generated by systematically varying key physical parameters, including Prandtl number, Casson parameter, stratification effects and permeability. A DARX-NNs-LMT is used to model the system. The model performance is evaluated through training, validation and testing phases using error analysis, regression plots and mean squared error to ensure accuracy and convergence. Findings The results of this study demonstrate that the proposed DARX-NNs-LMT model achieves excellent agreement with reference solutions, with errors ranging from 10–2 to 10–9, confirming its accuracy and robustness. Convergence analysis through mean squared error, regression and error histograms validates the predictive capability of the model. This study reveals that increasing the material (micropolar) parameter significantly enhances both the velocity and micro-rotation profiles. Furthermore, variations in stratification, permeability and thermal parameters exhibit notable influences on heat and mass transfer characteristics, highlighting the effectiveness of the proposed computational framework in capturing complex fluid behavior. Research limitations/implications The model relies on simulated data sets and does not incorporate experimental validation. Additionally, the analysis assumes constant physical properties and neglects three-dimensional and time-dependent effects. Despite these limitations, this study provides a reliable computational framework, offering significant implications for extending the model to more complex geometries, variable properties and real-life applications in fluid dynamics and engineering systems. Practical implications The proposed DARX-NNs-LMT framework provides an efficient and accurate tool for predicting heat and mass transfer in complex non-Newtonian fluid systems. In biomedical engineering, it can assist in understanding blood flow behavior under varying thermal and compositional conditions. In industrial applications, the model offers practical value in optimizing polymer processing, chemical transport and thermal management processes. Its computational efficiency reduces reliance on costly numerical simulations, enabling faster design and analysis. The approach can be extended to support real-time monitoring and control in engineering systems involving stratified fluid flows. Social implications This study contributes to societal well-being by advancing the understanding of complex fluid behaviors relevant to biomedical applications, particularly blood flow and related physiological processes. Improved modeling of heat and mass transfer in such systems can support better diagnosis, treatment planning and medical device design. Additionally, the optimization of industrial processes enhances energy efficiency and reduces resource consumption, contributing to environmental sustainability. The integration of artificial intelligence in modeling promotes technological innovation, supporting the development of smarter and more efficient engineering solutions with broader societal benefits. Originality/value This study presents a novel integration of doubly stratified micropolar-CF modeling with a DARX-NNs-LMT. Unlike conventional numerical approaches, the proposed framework efficiently captures complex nonlinear dynamics with high accuracy and fast convergence. The simultaneous consideration of multiple physical effects, including thermal and solutal stratification, micropolarity and porous media, enhances the model’s realism. This work offers a valuable computational paradigm that bridges advanced fluid dynamics and artificial intelligence, providing a reliable and scalable approach for solving complex engineering problems.