Prediction of average daily gain of swine based on machine learning

Issue title: The 6th International Multi-Conference on Engineering and Technology Innovation 2017 (IMETI2017)

Guest editors: Wen-Hsiang Hsieh

Article type: Research Article

Authors: Lee, Woongsup^a | Han, Kang-Hwi^a | Kim, Hyeon Tae^b | Choi, Heechul^c | Ham, Younghwa^d | Ban, Tae-Won^{a; *}

Affiliations: [a] Department of Information and Communication Engineering, Institute of Marine Industry, Gyeongsang National University, Tongyeong, Republic of Korea | [b] Department of Bio-Industrial Machinery Engineering, Institute of Agriculture & Life Science, Gyeongsang National University, Jinju, Gyeongnam, Republic of Korea | [c] Livestock environment division, National Institue of Animal Science, Kongjwipatjwi, lseo, Wanju, JeonBuk, Republic of Korea | [d] Agrirobotech Co., Ltd., Sina-ro, Bubal-eup, Icheon-si, Gyeonggi-do, Republic of Korea

Correspondence: [*] Corresponding author. Tae-Won Ban, Department of Information and Communication Engineering, Institute of Marine Industry, Gyeongsang National University, 445, Inpyeong-dong, Tongyeong, Gyeongnam, 650-160, Republic of Korea. E-mail: twban35@gnu.ac.kr.

Abstract: Understanding factors affecting growth rates in swine is important in the productivity of pig farms. We herein propose machine learning-based schemes to predict the average daily gain (ADG) of pig weight using temperature, humidity, feed intake, and the current weight of the pig. In order to address the lack of available growth data for pigs, we generate a synthetic dataset describing the weight of swine in relation to environmental factors based on the theoretical growth model and experimentally measured data, in an attempt to facilitate the application of machine learning techniques. Using the generated growth data, linear regression, tree regression, adaptive boosting (AdaBoost), and a deep neural network (DNN) are applied to estimate ADG. By means of a performance evaluation, we confirm that the machine learning algorithms are capable of predicting the ADG of swine accurately even when the growth characteristics of pigs are heterogeneous, i.e., each pig follows a different growth curve. Moreover, we also find that DNN can provide a higher predictive accuracy than other machine learning-based schemes.

Keywords: Average daily gain, swine, machine learning, prediction, deep learning

DOI: 10.3233/JIFS-169869

Journal: Journal of Intelligent & Fuzzy Systems, vol. 36, no. 2, pp. 923-933, 2019