NJmatWeb: An AI-Driven Web Platform Automating Machine Learning and Symbolic Regression for Materials Science

Data-driven methodologies are now widely employed for predicting materials properties; however, most existing tools fail to integrate automated feature selection, interpretable modeling, and an accessible user interface into a unified pipeline. To bridge this gap, we introduce NJmatWeb, a web-based platform that automates machine learning (ML), symbolic regression and SHAP analysis for materials science applications. Designed to perform robustly on the small datasets commonly encountered in this field, NJmatWeb incorporates a symbolic regression module that yields interpretable symbolic expressions, thereby enhancing model transparency. NJmatWeb incorporates an automatic featurizer that handles both inorganic and organic materials. The platform further provides automated feature selection, feature ranking, SHAP analysis, dataset visualization, and heat map comparisons of the feature space before and after selection. Prediction accuracies of available models are reported, with results from both widely adopted materials benchmarks and diverse case studies. This study demonstrates a useful AI tool aligned with the Materials Genome Engineering, accelerating materials design by combining predictive power with interpretability in an accessible web platform.