Why progress in multi-objective materials design via machine learning depends on integrated data infrastructure

Machine learning is transforming materials discovery, yet its impact on multi-objective optimisation remains constrained by the structure and quality of input data. Despite the growing use of advanced predictive models, most studies rely on single-property datasets developed for specific purposes and not designed for cross-domain integration. Fragmented descriptors, inconsistent metadata, and missing values prevent models from learning relationships across properties and material classes, limiting effective optimisation. This perspective focuses on data-driven materials informatics (with an emphasis on metal alloys) and contends that progress in multi-objective materials design relies on integrated, reliable data infrastructures that enable cross-domain learning and support practical optimisation. We demonstrate this by unifying fragmented mechanical and corrosion databases into a single resource for multi-objective design of multi-principal element alloys. Machine learning promises accelerated materials discovery, but fragmented and inconsistent datasets continue to limit effective multi-objective optimisation across materials systems. In this Perspective, the authors argue that integrated data infrastructures are essential for cross-domain learning and demonstrate this by unifying mechanical and corrosion databases to enable the multi-objective design of multi-principal element alloys.