Leveraging advanced ensemble learning techniques for methane uptake prediction in metal organic frameworks.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Adsorbed Natural Gas (ANG) technology requires efficient adsorbent materials for methane storage.
• Metal-Organic Frameworks (MOFs) show promise due to their high surface area and tunable properties.
• Accurately predicting methane adsorption in MOFs presents significant computational and experimental challenges.

Purpose of the Study:

• To implement ensemble machine learning (ML) models for accurate estimation of methane uptake capacity in MOFs.
• To evaluate gradient boosting variants (GBoost, XGBoost, LightGBM, CatBoost) for predicting MOF methane adsorption.
• To establish a predictive model using readily available MOF features and experimental conditions.

Main Methods:

• Developed a database of approximately 2600 data points for experimentally synthesized MOFs.
• Applied ensemble-based ML algorithms, including XGBoost, to predict methane uptake.
• Utilized features such as temperature, pressure, pore volume, and surface area.
• Implemented outlier detection to ensure data validity.

Main Results:

• XGBoost demonstrated superior performance with a correlation coefficient (R²) of 0.9955.
• The model accurately predicted the physical trend of methane (CH₄) capacity variations with pressure.
• Pressure was identified as a highly impactful feature in the predictive model.
• Outlier analysis confirmed that approximately 95% of the data points were valid.

Conclusions:

• Ensemble ML, particularly XGBoost, provides a highly accurate method for estimating methane uptake in MOFs.
• The developed model can reliably predict methane adsorption behavior based on key MOF properties and conditions.
• This approach facilitates the screening and development of advanced MOF materials for ANG applications.