Dataset and code for machine learning prediction of effective thermal conductivity of a hexagonal metal-matrix composite

This record contains the dataset and source code used in the study on machine learning prediction of the effective thermal conductivity of a metal-matrix composite with hexagonal fiber packing. The dataset was generated using finite element modeling in ANSYS APDL. It includes 1000 numerical experiments with input parameters describing the relative fiber radius and thermal conductivities of the fiber and matrix. The target variables are the effective thermal conductivity coefficients Ky and Kz. The record also includes the ANSYS APDL script, Python scripts for training and evaluating a multilayer perceptron model, GroupKFold validation metrics, and a neural network architecture figure.