Machine learning offers a promising avenue for expediting the discovery of new compounds by accurately predicting their thermodynamic stability. This approach provides significant advantages in terms of time and resource efficiency compared to traditional experimental and modeling methods. However, most existing models are constructed based on specific domain knowledge, potentially introducing biases that impact their performance. Here, we propose a machine learning framework rooted in electron configuration, further enhanced through stack generalization with two additional models grounded in diverse domain knowledge. Experimental results validate the efficacy of our model in accurately predicting the stability of compounds, achieving an Area Under the Curve score of 0.988. Notably, our model demonstrates exceptional efficiency in sample utilization, requiring only one-seventh of the data used by existing models to achieve the same performance. To underscore the versatility of our approach, we present three illustrative examples showcasing its effectiveness in navigating unexplored composition space. We present two case studies to demonstrate that our method can facilitate the exploration of new two-dimensional wide bandgap semiconductors and double perovskite oxides. Validation results from first-principles calculations indicate that our method demonstrates remarkable accuracy in correctly identifying stable compounds.
© 2024. The Author(s).