Machine Learning–Assisted Design of Ytterbium-Based Materials with Tunable Bandgaps and Enhanced Stability

Abstract

This study presents a comprehensive machine learning (ML) investigation into the design of ytterbium (Yb)–based materials with tunable bandgaps and enhanced stability. A dataset of 2062 Yb-based compounds was compiled from the literature, featuring structural and electronic properties. Their descriptors are designed to calculate their bandgap values through quantum chemical studies. Out of various machine learning models, Random Forest and Decision Tree regression models produce the more accurate prediction for their predicted values. The feature importance analysis reveals that their HeavyAtomCount, AliphaticRings, Ar_Rings, and TPSA emerge as important descriptors to influence their bandgap and stability. Their stability analysis by their Convex Hull Diagrams identifies 73% of data within its stability boundary. Further insights are gained through their clustering analysis, utilizing K-means clustering (k = 4), t-SNE, and elbow methods. The results show 85% accuracy to predict their stability and reveal distinct their chemical profiles as stable, moderately stable, and less stable compounds. The current study design approach is supposed to enable the discovery of Yb-based compounds with their tailored properties, like their tunable bandgaps (0.5–4.5 eV) with their enhanced stability (> 70%). The study can provide insights for materials scientists to identify promising optoelectronic materials.