A novel data-driven framework integrating physical feature engineering, interpretable machine learning and symbolic regression techniques is suggested to establish a complete mapping from polymer micro configurations to thermal conductivities, as well as the efficient identification of high thermal conductivity polymers. A team led by Prof. Shenghong Ju from the China-UK Low Carbon College, Shanghai Jiao Tong University, China, proposed a physical feature engineering for representing complex polymer systems, which combines multiscale simulations and interpretable machine learning for establishing an informatics model on the relationship between the microstructures and thermal conductivity in polymer systems, and reveals a unified physical rule of the hierarchical structures and thermal conductivity: (a) High thermal conductivity polymer chains usually have strong intra-chain interatomic interactions, and the corresponding amorphous polymers are prone to exhibit improved thermal conductivity. (b) Most of the identified high thermal conductivity polymers are conjugated, facilitating the maintenance of a large chain stiffness. (c) Since polymer chains have a highly aligned chain orientation, their thermal conductivity variations are more sensitive to changes in chain ordering due to dihedral rotation. The proposed approach may assist in the research of high-performance polymers that are not limited to thermal conductivity, and aid in understanding the linkage between the properties of different hierarchical structures.