Machine learning models for biomass energy content prediction: A correlation-based optimal feature selection approach

Abstract

In this study, a multilinear regression (MLR) and three machine learning techniques, i.e., an adaptive neuro-fuzzy inference system (ANFIS), an artificial neural network (ANN), and a support vector machine (SVM) were employed to develop biomass higher heating value (HHV) prediction models as a function of the proximate analysis. Seven inputs selection were applied to explore the extent of correlation between the independent variables and the HHV. The pairing of the volatile matter and fixed carbon presented the most accurate model in ANN, SVM, and MLR while in ANFIS, the ash combined with fixed carbon was more effective. Overall, the combination of ash and fixed carbon in ANFIS was superior in prediction performance having presented the highest correlation coefficient of 0.9371 and the least mean squared error of 0.0029. These techniques can guarantee precise predictions of the HHV of biomass using proximate analysis instead of rigorous and expensive experimental procedures.