Estimation of BMS using image analysis traits with Random Forest in Japanese Black cattle

MIYATA Ayu, KUCHIDA Keigo

doi:10.2508/chikusan.94.295

<p>Recently, the use of artificial intelligence in animal science field has garnered attention. Beef marbling standard (BMS) is one of the most important carcass traits for determining the value of carcasses. This study aimed to estimate BMS using information obtained from carcass cross sectional images at the 6th and 7th rib with Random Forest, a type of machine learning. Rib eye images of 2,498 Japanese Black cattle that were shipped to one market in Hokkaido between January and December 2022 were analyzed. Twenty-three image analysis traits gained in the rib eye were used as explanatory variables. Stratified k-fold cross-validation with k＝5 was performed to evaluate the estimation performance of Random Forest, including the difference between grading BMS and estimated BMS, the importance of variables, and Shapley values. The percentages of difference between grading BMS and estimated BMS within ±0, ±1, and ±2 were 51.8%, 94.1%, and 99.7%, respectively. The importance of each variable, in descending order, was 0.8634 for marbling percentage, 0.0297 for new fineness index, and 0.0121 for coarseness index 1-10. The results showed that the amount and shape of marbling were particularly important for estimating BMS. The amount and fineness of marbling had positive effects on estimates, while the coarseness of marbling acted as a negative factor, especially in intermediate BMS (7 to 10). Therefore, using Random Forest has suggested that highly accurate and interpretable estimation of BMS predictions is feasible.</p>

Estimation of BMS using image analysis traits with Random Forest in Japanese Black cattle

Bibliographic Information

Search this article

Description

Journal

References(15)*help

Keywords

Details 詳細情報について

Export

Report a problem