DATA-DRIVEN DISCOVERY OF BORON CARBIDE-LIKE MATERIALS USING OPEN COMPUTATIONAL MATERIALS DATABASES

Abstract

This study presents a data-driven approach for identifying materials with similar properties to target compounds using open computational materials databases. We use boron carbide (B₄C) with specific boron-to-carbon ratios (0.78-0.91) as our test case material and develop a robust pipeline for data collection, processing, and analysis. Six clustering algorithms representing different methodologies and several feature engineering techniques are employed to group materials based on structural, energetic, and mechanical properties. Our methodology successfully identifies clusters with high concentrations of target boron carbide compositions and reveals other boron-containing materials with similar properties. This approach demonstrates the potential of leveraging open materials databases for accelerated materials discovery and provides a framework applicable to various material systems beyond boron carbide.