Predictive Modeling of Higher Heating Value of Biomass Using Ensemble Machine Learning Approach

Higher heating value (HHV) of biochars serves as a critical and vital component for the determination of biomass economy. The complex biomass structure with time-consuming and costly experiment set-up necessitates HHV estimation using various machine learning approaches. In this work, a large dataset of 1140 data samples based on proximate analysis were analyzed using combinational approach of grade and value prediction for the HHV estimation. Three ensemble machine learning algorithms, namely: (i) Bagging, (ii) Multiclass classifier (MCC) and (iii) Classification-via-regression (CVR), were used in pair with the random forest (RF) and multilayer perceptron (MLP) models for the prediction of HHV. Parameter optimization and model formation using meta-classifiers improved the prediction accuracy of the designed models. Bagging and CVR meta-classifier paired with RF led to the highest sensitivity values (0.983 and 0.982, respectively) and lowest specificity values (0.0008 and 0.0009, respectively) with respect to all other designed models. RF model showed highest correlation coefficient (CC) value of 0.984 and lowest root-mean-square (RMSE) value of 1.4204.