Use of an Artificial Neural Network and Multiple Linear Regression for the Prediction of the Asymptotic Gas Production of Agricultural By-products

This study evaluated the correlation between chemical composition and asymptotic in vitro gas production (AGP) of various agricultural by-products. In addition, it attempted to use artificial neural network (ANN) and multiple linear regression (MLR) approaches to predict AGP based on the chemical composition of these by-products. Two data sets were used in the study. The first dataset (training) consisted of published data, while the second dataset (testing) consisted of the chemical composition and asymptotic gas production of selected by-products. First, the chemical composition and AGP of the selected by-products were measured. The two data sets were then pooled and processed for cluster analysis. Multivariate cluster analysis revealed two distinct groups (A & B) with a degree of similarity among the by-products within each group exceeding 80% and 90%, respectively. The selected by-products were categorized into cluster A, so this cluster was considered for Pearson correlation and modeling analysis. Negative correlations were observed between AGP and NDF and ADF. Conversely, positive correlations were found between OM and CP of agricultural by-products and AGP. The ANN showed superior performance in predicting asymptotic gas production (b) compared to the MLR model, as indicated by lower root mean square error (RMSE) and higher r ² .