Integrating Experimental Insights with Machine Learning for Inverse Design of Sustainable Lignin-Derived Cr (VI) Adsorbents

s Lignin-derived carbonaceous materials have emerged as promising candidates for heavy metal remediation owing to their low cost, tunable surface chemistry, and hierarchical porosity. In this study, a FeCl 3 –urea co-activation strategy was employed to synthesize N-doped hierarchical porous carbons (N-HPCs) from lignin for efficient Cr (VI) removal. The synthesis parameters were systematically optimized using a machine learning (ML) workflow, with experimental data serving as the training set. Gradient boosting regression (GBR) achieved high predictive accuracy (R 2 = 0.86), and SHAP-based feature importance analysis identified pyridinic-N (N-5) content as the dominant factor governing Cr (VI) adsorption, attributed to its redox-active nature. Under optimal conditions (FeCl 3 :lignin:urea = 3:1:0.5), the resulting N-HPC exhibited a high specific surface area (2297.5 m 2 /g), elevated N-5 content (67.08%), and a maximum Cr (VI) adsorption capacity of 627.1   mg/g. Mechanistic studies revealed that reduction–complexation, primarily facilitated by N-5 sites, accounted for up to 87.7% of Cr (VI) removal. The ML-driven inverse design strategy further predicted that precise control of N-5 levels could yield N-HPCs with adsorption capacities exceeding 650   mg/g. This work demonstrates the synergy of ML-guided optimization and mechanistic insight, providing a robust framework for the rational design and accelerated development of sustainable, high-performance adsorbents for heavy metal remediation.