Applicability of statistical and machine learning-based regression algorithms in modeling of carbon dioxide emission in experimental pig barns

Abstract

Pig farming is one of the major sources of greenhouse gas (GHG) emissions in the agricultural sector; nevertheless, few studies have been undertaken to directly measure or estimate GHGs, particularly carbon dioxide (CO2) from pig barns. Therefore, the main objective of the present research was to estimate and predict CO2 emission rate as a function of the mass of pigs and feed consumption. Two identical experiments were carried out in experimental pig barns in 2020 and 2021 to develop and evaluate the performance of CO2 emission model. The CO2 emission data (ppm) were collected utilizing Livestock Environment Management Systems (LEMS) and weather sensors, respectively within the pig barns and the outside environment. The models were built using seven statistical and machine learning-based regression algorithms, i.e., linear, multiple linear, polynomial, exponential, ridge, lasso, and elastic net. The findings of the study revealed that among the seven models, the exponential-based regression model performed the best, with a coefficient of determination (R-2) greater than 0.78 in the training stage and 0.75 in the testing stage being suitable to describe the relationship between the feed intake and the rate of CO2 emission. However, when compared to the other models in the testing stage, the lasso model had the worst performance (R-2 < 0.65 and RMSE > 20.00 ppm). In conclusion, this study recommends employing an exponential-based regression model by taking feed intake as an input variable in predicting CO2 for a small number of the experimental dataset.