Random forrest regression (RFr) data upscaling and artificial neural network (ANN) modeling for sinusoidal microchannel heatsinks with graphene nanofluids

• Amplitude increased heat transfer up to +5.9% in sinusoidal microchannels. • High ω–A coupling reduced Nu by up to −2.7%. • Nanoparticles raised Nu more at low Re (+2.9%) than high Re (+1.4%). • RFr cut MSE by 58.4% and improved R by 8.8%. • ANN predictions validated with <5% average error (min 0.07%). Predicting heat transfer in sinusoidal microchannel heat sinks is challenging because numerical simulations are computationally expensive and the resulting datasets remain sparse across the design space. This study integrates high-fidelity CFD with a two-step hybrid machine learning framework to model graphene nanofluid flow in sinusoidal microchannels. Key parameters include Re, nanoparticle volume fraction, waveform amplitude, frequency, width, and number of segments. The results show that amplitude enhances heat transfer by up to 5.9%, although high frequency and amplitude combinations reduce the average Nusselt number by 2.7%. Nanoparticle effects are more pronounced at low Reynolds numbers, and geometric variations, such as increasing the width, reduce the Nusselt number. To overcome data sparsity, Random Forest regression is used to upscale the numerical dataset before training an artificial neural network surrogate model, reducing mean squared error by 58.4% and improving regression by 8.8%. Validation against additional CFD cases yields an average prediction error below 5%. The proposed framework offers a computationally efficient and scalable tool for microchannel heat-sink design.