Machine learning algorithms in predicting piglet mortality and weight based on sow backfat thickness
Year: 2025
Type: article
Abstract: The implementation of machine learning models to predict reproductive variables in sows represents a fundamental tool for optimizing management in swine production systems. The objective of this study was to evaluate the predictive capacity of random forest (RF) and dense neural network (DNN) models to predict live birth piglets (LBP), total piglets born (TPB), average piglet weight (APW), and piglet mortality rate (PMR), using backfat thickness (BFT) as the predictor variable. A total of 571 records from a commercial swine farm were analyzed, of which 348 were used after data cleaning. Models were developed in Google Colaboratory using cross-validation for hyperparameter tuning. Performance was evaluated using mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE), and relative percentage error (RPE). For LBP, MAE was 1.51 in both models, MSE of 3.62 (DNN) and 3.53 (RF), RMSE of 1.90 (DNN) and 1.88 (RF), with RPE of 13.59% and 13.41%, respectively. For TPB, the MAE was 1.61 in both models, with MSE values of 3.99 (DNN) and 3.95 (RF), and RMSE values of 2.00 (DNN) and 1.99 (RF), respectively. The RPE was 13.74% and 13.64%, respectively. For APW, both models showed satisfactory fit with RMSE of 0.063 kg and RPE close to 4.5%. PMR prediction was poor, with RPE exceeding 92% in both models. RF and DNN models demonstrated predictive potential for LBP, TPB, and APW using only BFT.
Authors: William Armando Tapie, Carlos-Leonardo Guerra-Marín, Juan Fernando Manrique-Hincapie, Sebastian Montoya Uribe
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
SDG: Zero hunger
Open Access status: green
OpenAlex ID: W7106369475