Search

Feature engineering remains central to materials informatics, yet systematically introduces scientific blind spots that constrain discovery and interpretation. These blind spots arise from choices in descriptor selection, transformation, and dimensionality reduction that inadvertently prioritize statistical correlations over physical invariance, overlook multi-scale interactions, and embed dataset-specific biases into model architectures. In small-data regimes common to materials science, engineered features often amplify overfitting while diminishing generalizability across chemical spaces. Interpretability suffers as complex engineered descriptors obscure mechanistic linkages between atomic structure and macroscopic properties. Literature consistently highlights these limitations across perovskites, alloys, energy materials, and porous systems, underscoring the tension between predictive performance and scientific fidelity. This conceptual manuscript synthesizes these challenges and proposes an original Integrated Blind Spot Navigation Model (IBSNM). The framework organizes feature engineering around four interdependent pillars—physical consistency guardrails, multi-scale descriptor integration, uncertainty-aware selection, and iterative co-interpretation—linked by feedback mechanisms that surface and mitigate hidden assumptions. By reframing feature engineering as a navigable landscape rather than a static preprocessing step, the model offers a conceptual pathway toward more robust, transparent materials informatics practices that do not rely on empirical validation.

This review systematically maps the scientific blind spots in the materials artificial intelligence literature by conducting a targeted search across key databases and journals to identify both what is heavily studied and what remains systematically invisible. The analysis organizes these blind spots into five interconnected categories—data, methods, evaluation, epistemic, and social—drawing on core peer-reviewed publications that collectively document the selective lens through which the field presents its progress. Key findings reveal that data blind spots center on underrepresented chemistries, structures, and operational conditions that leave critical real-world properties unmodeled; methodological blind spots arise from the dominance of correlation-driven approaches while uncertainty quantification, small-data techniques, and causal methods receive scant attention; evaluation blind spots manifest in the near-total absence of distribution-shift testing, robustness checks, and negative-result reporting that inflate perceived reliability; epistemic blind spots persist through prediction-without-explanation paradigms and the failure to articulate model boundary conditions or failure modes; and social blind spots ignore value-laden assumptions, equity considerations, and the broader societal and environmental implications of materials AI deployment. Synthesis across categories uncovers systemic patterns such as positive publication bias creating self-reinforcing feedback loops, methodological innovation consistently outpacing rigorous evaluation, data gaps mirroring decades-old research priorities, and epistemic shortcomings that propagate through every layer of the pipeline. Recommendations, therefore, target authors, reviewers, journals, and funders with concrete actions to surface these blind spots, thereby enabling a more balanced, reproducible, and societally relevant materials AI research agenda that closes the gap between published claims and real-world impact.