Validation, improvement, and data mining of machine learning predictive models for large white male turkey body weight

Authors: Karlinton R. Flores, Benjamin R. Reading, Jesse L. Grimes

Ger. J. Vet. Res 2025. vol. 5, Iss. 3 pp:42-58
Doi: https://doi.org/10.51585/gjvr.2025.3.0147

Download Options

×

• Download PDF
• Download XML
• Download Figures
• Supplementary Files

Close

Abstract:

Validation of predictive models is as critical as their development, and model improvement relies on the iterative incorporation of additional data. This process, however, can be costly, as experiments must adhere to consistent designs. Machine learning (ML) facilitates model validation by evaluating standard variables across data sets that were previously analyzed in isolation. Furthermore, ML enables the integration of new data into existing models and supports data mining (DM) to uncover novel patterns. The present study demonstrates the application of ML to validate, refine, and extend previously developed classification and regression models predicting body weight (BW) in Large White male turkeys. Performance, environmental, and microbiome data from two prior experiments, along with the original data set used for model construction, were analyzed. In total, 279 variables, including BW, gut pH, cloacal temperature, corn particle size, mineral source, genetic line, and microbiome taxa, were processed through ML and DM analyses. The Waikato Environment for Knowledge Analysis (WEKA) 3.8.5 Experimenter tool was used to test multiple classification and regression algorithms, employing 10-fold crossvalidation, to predict BW at 18 weeks. Model validation across the three data sets yielded a correlation coefficient of 0.59 and a root mean square error (RMSE) of 1.03. Among the tested algorithms, M5' generated the most interpretable and practical model, achieving a correlation of 0.81 and an RMSE of 0.58 across four constructed models. Key predictors of 18-week BW included genetic line, rearing season, and BW at 5 and 14 weeks, while twelve microbiome taxa also contributed significantly to prediction accuracy. In conclusion, ML-based BW prediction models were successfully validated, improved, and expanded through DM approaches. Future research should focus on incorporating larger and more diverse data sets to enhance model robustness and predictive power.

Keywords:

Turkey production, Machine learning, Data mining, Turkey performance

Statistics:

Article Views: 547
PDF Download: 6