Hanwoo Carcass Dataset for Backfat Thickness Modeling and Machine Learning Benchmarking
Year: 2026
Type: dataset
Abstract: Overview This dataset accompanies the study entitled "Comparative benchmark of machine learning models for modeling backfat thickness in Hanwoo cattle under nested cross-validation." The dataset comprises 386 animal-level carcass records from Hanwoo (Korean native cattle) collected between 2007 and 2020. The records were obtained from eight previously completed research projects conducted using data provided by Jeonbuk National University and the National Institute of Animal Science, Republic of Korea. The dataset was prepared to support comparative benchmarking of machine learning regression algorithms for modeling backfat thickness using slaughter-record and carcass-related phenotypic variables. Dataset Description The original pooled database contained 16,837 sample-level records from 520 animals. One record per animal was retained, and animals with missing required measurements were excluded, resulting in 386 records. The final dataset comprises 37 bulls, 245 steers, and 104 cows/heifers, including hysterectomized heifers. Backfat thickness (mm) is the target variable. Nine predictors were retained: Month: Age at slaughter (months) FatNo: Intramuscular fat score ColorNo: Meat color score Fatcol: Subcutaneous fat color score Texture: Meat texture score Maturity: Physiological maturity score agdays: Meat aging duration (days) WBsf: Warner–Bratzler shear force (kg) Sex: Animal sex category The quality-grade and quantity-grade scores were excluded from the primary predictor set to reduce redundancy and avoid potentially circular prediction. Research Application The accompanying study compared 14 machine learning regression algorithms, including linear and regularized regression, support vector regression, k-nearest neighbors, decision trees, ensemble learning methods, and multilayer perceptron models. Model performance was evaluated using nested cross-validation with five outer folds and three inner folds. Mean absolute error (MAE), root mean squared error (RMSE), and the coefficient of determination (R²) were used as evaluation metrics. This dataset is intended to support reproducible research, methodological benchmarking, and further investigation of phenotypic relationships among Hanwoo carcass traits. Important Considerations The dataset represents retrospective slaughter-record, carcass-grading, and meat-quality measurements. Aging duration and Warner–Bratzler shear force measurements correspond to the first recorded meat sample for each animal, and the sampled cuts and aging times varied among research projects. The dataset should not be interpreted as a prospective pre-slaughter prediction dataset or as evidence of causal relationships between the recorded variables and backfat thickness. Data Sources and Acknowledgments The authors acknowledge Jeonbuk National University and the National Institute of Animal Science, Republic of Korea, for providing access to the original carcass data. This research was supported by the Rural Development Administration, Republic of Korea, through the Cooperative Research Program for Agriculture Science & Technology Development (Project No. PJ0170202022).
Authors: Yun, Taeyong, Jeong, Dawoon, Lee, Woongsup
Institutions: Yonsei University, Jeonbuk National University
Language: English
Subfield: Genetics
Domain: Life Sciences
SDG: Zero hunger
Open Access status: green
OpenAlex ID: W7221661375