Hierarchical Prediction of Carcass Traits Using the Group Pliable Lasso
Conference
10th International Conference on Agricultural Statistics
Format: CPS Abstract - ICAS 2026
Keywords: carcass weight, group pliable lasso, marbling score, meat industry
Abstract
Reliable prediction of carcass traits is central to supporting decision-making in the red meat industry. In this study, conducted with a large commercial meat processor, we develop a hierarchical predictive framework for three key outcomes: dentition, marbling score, and carcass weight. Our approach is based on the group pliable Lasso, a penalised regression model designed to improve predictive accuracy while capturing context-specific effects. Unlike conventional models that assume uniform relationships, the pliable Lasso allows effects (e.g., lairage time) to vary with modifying factors such as abattoir, season, or temperature. It automatically discovers when and where effects differ, and keeps only the differences that truly improve predictions.
The framework operates in three stages: dentition modelling, prediction of marbling score using dentition outputs, and prediction of carcass weight using both dentition and marbling score. This layered design reflects biological and operational dependencies among carcass traits. Empirical results show that the group pliable Lasso provides accurate and interpretable predictions, outperforming standard regression benchmarks. In contrast to black-box machine learning approaches, it not only identifies which factors matter most but also reveals the specific conditions under which they matter (e.g., by plant or season).
This methodology highlights the value of structured penalised regression in the meat industry, complementing broader forecasting strategies and offering processors transparent tools for both prediction and explanation.