The accurate prediction of gas adsorption properties in metal-organic frameworks (MOFs) is crucial for applications in gas storage, separation, and sensing. Traditional computational methods, such as Grand Canonical Monte Carlo (GCMC), provide valuable insights but are often computationally expensive and infeasible under certain operating conditions. Machine learning (ML) approaches have emerged as efficient alternatives; however, they frequently struggle with limited training data and model complexity. In this work, we propose a recommendation system based on the MissForest algorithm to predict missing adsorption properties and address the cold start problem in MOF gas adsorption datasets. Our results show that MissForest performs well, achieving a normalized root mean square error (NRMSE) of 0.16 with only 20% of the data observed. This framework offers a scalable and data-efficient approach for guiding gas adsorption predictions.
Gbenga Fabusola is a 4th year PhD student in Dr. Cory Simon’s lab at Oregon State University. His research focuses on applied machine learning for gas sensing technology. Outside of professional settings, Gbenga loves hitting the gym and playing football.