Machine learning-guided design/optimisation of functionalized metal–organic frameworks for sustainable carbon capture and photocatalytic conversion
Abstract
Achieving Sustainable Development Goal 13 necessitates a rapid reduction of anthropogenic CO₂ emissions to limit global temperature rise below 1.5 °C. Carbon capture and storage (CCS) technologies—such as post-combustion, oxy-fuel, and membrane-based processes provide potential mitigation strategies but are often constrained by energy costs and scalability issues. Metal organic frameworks (MOFs) have gained attention for CO₂ adsorption and photocatalytic conversion due to their large surface area, tunable porosity, and adjustable electronic structures. Surface functionalization and heterostructure formation further enhance charge separation and improve photocatalytic efficiency. A comprehensive dataset of 6001 MOFs derived from the publicly available QMOF database, encompassing structural, electronic, and synthesis descriptors, was analysed using four machine learning models. Among these, the multilayer perceptron (MLP) achieved a test R² of 0.9567 and MAE of 6.96, with five-fold cross-validation confirming robust generalisation (CV R² = 0.9585 ± 0.0018). SHAP analysis identified sacrificial agent selection, calcination temperature, and bandgap as the dominant performance indicators. This integration of machine learning with materials design enables data-driven discovery and optimisation of MOFs for efficient CO₂ capture and conversion applications.