Estimating biomass major chemical constituents from ultimate analysis using a random forest model

Area of Science:

• Biomass characterization
• Computational chemistry
• Sustainable energy

Background:

• Accurate determination of biomass chemical constituents is crucial for its effective utilization.
• Experimental methods for chemical analysis are often costly and time-consuming.
• Existing predictive models have limitations in accuracy and applicability.

Purpose of the Study:

• To develop a novel random forest (RF) model for predicting major biomass chemical constituents.
• To compare the RF model's performance against traditional correlations and experimental data.
• To provide a more efficient and accurate method for biomass characterization.

Main Methods:

• Construction of two databases from existing literature for RF model training and application.
• Development and implementation of a random forest regression model.
• Validation using determination coefficients (R²) and mean absolute percentage error (MAPE).

Main Results:

• High determination coefficients (R²) achieved during training: 0.954 (cellulose), 0.933 (hemicellulose), and 0.968 (lignin).
• Accurate predictions on diverse biomass samples with MAPE < 20% and R² values of 0.862 (cellulose), 0.904 (hemicellulose), and 0.962 (lignin).
• Outperformed previous correlations, which showed high errors (MAPE > 500%) and unrealistic predictions for external samples.

Conclusions:

• The developed RF model offers a highly accurate and reliable approach for predicting biomass chemical constituents.
• This computational method significantly reduces the time and cost associated with biomass analysis.
• The RF model demonstrates broad applicability across various biomass types, surpassing limitations of prior methods.