Meat yield prediction of the red swamp crayfish Procambarus clarkii based on machine learning

Meat yield is a complex and economically significant trait for the red swamp crayfish Procambarus clarkii . Identifying the most suitable weight range for market readiness and establishing an integrated machine learning model to predict the meat yield addresses widespread challenges in P . clarkii , holding economic value and practical importance for the breeding of high-yield P . clarkii . Phenotypic traits including full length (FL), body length (BL), abdominal length (AL), the second uromere length (2UL), width (2UW), and height (2UH), along with body weight (BW), were measured from 1392 P. clarkii collected from cultivated paddy fields. Meat yield (MY) was calculated through dissection and weighing of tail meat weight (TMW). To assess the sexual dimorphism, inter-trait correlations, and multicollinearity, multiple descriptive statistical tests were employed on the collected dataset, including the data normalization, Pearson correlation coefficient calculation, and Variance Inflation Factor (VIF) calculations. To determine the optimal weight which optimizes MY and TMW across both sexes Lowess bivariate curves were utilized for fitting. various machine learning methods were utilized to predict the meat yield of P. clarkii , evaluated against an independent test dataset. Additionally, to understand how key traits contribute to the predictive probability of P. clarkii meat yield, SHapley Additive exPlanations (SHAP) interpretability methods were employed on the most accurate predictive algorithm. The AL, 2UW, 2UH, BW, and MY were statistically significant differences between sexes. FL, BL, AL, 2UL, 2UW, 2UH, BW, TMW were all highly positively correlated with each other, while MY was negatively correlated with these traits. The VIF test showed a high degree of collinearity among BL, FL, BW, and TMW. The optimal weight for maximum meat yield and tail meat was identified as 27.75 g for females and 17.51g for males. Across three machine learning models we conducted, the Gradient Boosting Regressor (GBR) model consistently produced the best predictions for meat yield in both sexes with. SHAP analysis revealed that the most critical trait examined was BW, with an explanation percentage of 40.8% in females and 46.6% in males. For females, FL (18.3%), 2UW (10.9%), and BL (10.7%) explained a high proportion among the traits included, whereas for males, BL (24.0%), AL (11.7%) and 2UH (6.6%) were more pronounced. In this study, an robust machine learning framework for accurately predicting P. clarkii meat yield was developed based on complex phenotypic traits. Compared to traditional destructive measurement methods and conventional regression models, this approach significantly enhances both operational efficiency and predictive accuracy. Additionally, the optimized weight threshold offers valuable insights into the commercial P. clarkii breeding industry, ensuring an optimal balance between growth performance and meat production efficiency.