Abstract
The previous drawback in spinel materials research is that screening for direct band gaps and thermal stability using conventional ab initio methods is computationally intensive and slow, limiting the discovery of promising candidates. In this research, a machine learning-based framework combining a Dense Neural Network (DNN) with a semiconductor classification model and the Extreme Gradient Boosting Algorithm (XGBA) is employed to accelerate ab initio predictions of spinel materials (AB 2 X 4 ). From a dataset of 4160 unexplored spinels, the model successfully identifies 185 candidates exhibiting direct band gaps and thermal stability. The framework achieves a prediction accuracy of 97.8% while significantly reducing computational costs and shortening the typical spinel screening research cycle by approximately 3.4 years. These results demonstrate the efficiency and high-throughput capability of the proposed strategy, offering a practical approach for the rapid discovery of novel spinel materials with desirable electro-optical and thermal properties.
Funding Information
  • Department of Science and Technology (DST), Ministry of Science and Technology, Government of India (DST/TMD/IC-MAP/2K20/03(G))

This publication has 53 references indexed in Scilit: