| Abstrakt: |
Previous studies have utilized traditional regression analysis to predict carcass unit prices. However, these studies have been limited in terms of the number of variables considered due to issues of collinearity. Additionally, no study has been conducted that treats both carcass grading traits and image analysis traits as variables without any restrictions. In contrast, Random Forest (RF) machine learning algorithm is well-studied for forecasting purposes and handle a larger number of variables simultaneously, without being affected by collinearity. Furthermore, RF can quickly and accurately learn regression analysis. In this study, the RF regression model was employed to predict carcass unit prices based on image analysis traits and carcass grading traits. The dataset used in the study consisted of 13,046 Japanese Black cattle carcasses that were shipped to meat processing plants in Hokkaido during 2015-2019 and 2022. Statistical analysis was involved using RF to predict carcass unit prices, with variables including carcass grading traits (18 traits), image analysis traits (16 traits), age, sex, farmer, sire, Tokyo market price (average carcass unit price of A4 castrated cattle) for the week prior to the previous day, traded month, order of entry, defect 1, and defect 2. The feature importance analysis revealed that meat quality, market price, and yield were the most influential factors in predicting carcass unit price, with all three factors having a positive impact. The prediction accuracy, as measured by the R2 value, was 0.86, indicating that the RF regression model can achieve higher accuracy in predicting carcass unit prices compared to conventional regression analysis. However, the results also indicate that it is challenging to accurately predict carcass unit prices for carcasses with a high degree of defects or for imbalanced carcasses. When using the model created with data from 2015 to 2019 to predict carcass unit prices in 2022, the accuracy decreased and the trends in determining carcass unit prices varied across different years. To improve the prediction accuracy, it is recommended to incorporate fatty acids, degree of defects, and objective values for the degree of marbling extending into the top and bottom round as additional variables. Furthermore, it is important to accumulate more years of training data in order to develop a robust model that can account for carcasses with diverse characteristics. If this method is implemented in the future, it is anticipated that the prediction of carcass unit prices will serve as a foundation for setting limit prices at auctions or for conducting relative transactions, ultimately leading to more efficient carcass transactions. [ABSTRACT FROM AUTHOR] |
