Generative Bayesian machine learning framework for inverse alloy designing of lightweight high entropy alloy with target density and experimental validation

Lightweight high entropy alloys (LHEAs) are promising next-generation structural materials due to their exceptional strength, ductility synergy, and thermal stability. The rational control of alloy density remains a key challenge due to the complex compositional design space. The presence of multiple principal elements introduces local chemical and structural heterogeneities that can produce measurable deviations from ideal rule-of-mixture density estimates. In the present work, a generative Bayesian machine learning (ML) framework is developed for the inverse design of lightweight HEA compositions with desired density targets. Physically informed descriptors derived from elemental properties, including atomic weight, valence-electron concentration, atomic size mismatch, melting temperature, mixing enthalpy, and electronegativity mismatch, are used to train supervised ML models. Bayesian optimised extra trees regression model is used to predict the alloy density with high accuracy, while providing uncertainty estimates through ensemble variance, and a complementary phase classification model ensures physically plausible phase stability. The trained ML models are unified into a generative inverse design workflow that proposes candidate LHEA compositions satisfying user-defined density constraints. Selected compositions are synthesised using the vacuum arc melting route, and their densities are measured experimentally to validate the model predictions. The combined computational-experimental approach enables a systematic assessment of model reliability and reveals the role of chemical and structural disorder in governing deviations from rule-of-mixture estimates of density. The present investigation demonstrates a practical pathway for machine learning-guided discovery and experimental validation of LHEAs, bridging generative Bayesian modelling and materials synthesis to accelerate alloy design.