Adsorption energy prediction and mechanistic analysis of metal atoms on MXene surfaces based on DFT and machine learning

MXenes show promising application prospects in metal-ion battery electrodes due to their advantages like adjustable surface chemical properties and open layered structure. However, the conventional density functional theory (DFT) method faces challenges of low efficiency and high time consumption when used for systematic exploration of the adsorption performance a large variety of MXenes. This study proposes an efficient prediction framework for adsorption energy based on machine learning. By integrating DFT data, a cross-element data set containing 90 types of MXene substrates and 5 types of metal atoms is constructed. We propose a new set of features to divide the MXene surface atoms into three layers according to their distance from metal atoms and extract the key features, such as electron affinity, the number of atoms closest to the metal atom, and atomic radii in each layer. Using eXtreme Gradient Boosting (XGB), Random Forest Regression (RFR), Neural Network (NN) and Support Vector Regression (SVR) models for comparative analysis, it is found that XGB shows the best performance on the test set (MAE = 0.21 eV, R² = 0.93). The SHAP interpretability analysis shows that the electron affinity of the first layer atoms and of the adsorbed metal atoms are the core factors affecting the adsorption energy. Finally, we select the MXenes with high adsorption energies, namely Sc₂NO₂ and Y₂NO₂ with high d band center, as prototypical examples for further analysis. Their d-band centers (2.0325 eV and -1.8770 eV) lie closer to the Fermi level, leading to higher adsorption energies. © 2026 The Authors.