DATA-DRIVEN PREDICTION OF PIGLETS BORN ALIVE PER FARROWING USING MONTE CARLO-OPTIMISED MULTIPLE REGRESSION

Resumen

In this study, a data-driven nonlinear power regression approach with Monte Carlo-based exponent optimisation was applied to predict the number of piglets born alive per farrowing (PFA). The analysis was based on productivity data from F1 sows obtained by crossbreeding Danish Landrace and Danish Large White breeds and inseminated with semen from Danish Duroc terminal line boars. Data were collected from 114 commercial farms in Denmark and divided into a primary dataset used for model development and an independent test dataset. The Monte Carlo random-search procedure enabled automatic identification of the optimal power exponents, thereby reducing the need for arbitrary variable transformations and improving model accuracy. The final model included a combined nonlinear predictor based on the number of weaned piglets per sow per year (PPSY) and piglet survival rate (PSR%), together with a quadratic term describing the average annual number of farrowings per sow (F). The model achieved high predictive accuracy, with R² = 0.99632, adjusted R² = 0.99623, RMSE = 0.04330, RRMSE = 0.23932%, and CCC = 0.99816. The test dataset RMSE was 0.04725, and the test-to-dataset RMSE ratio was 1.09, indicating good predictive stability and no substantial overfitting. The developed model demonstrated high predictive accuracy, statistical robustness, satisfactory diagnostic performance, and good generalisability. The results indicate that data-driven nonlinear power regression with Monte Carlo-based exponent optimisation may provide a useful approach for modelling biological and zootechnical data, particularly when nonlinear relationships are expected.