Smart Agricultural Technology (Oct 2023)
Comparing forecasting models for predicting nursery mortality under field conditions using regression and machine learning algorithms
Abstract
Swine nursery mortality is one of the most important key performance indicators (KPIs) for commercial swine production systems and, for the purpose of this study, refers to the number of pigs that died during the first 60 days during the post-weaning phase (approximately 170-day period). Likewise, nursery mortality accounts for almost 50% of the mortality that occurs throughout the post-weaning phase. Despite being a metric concerning the post-weaning phase of production, previous studies have demonstrated that nursery mortality is highly correlated with the previous performance of the piglets in the pre-weaning phase and to stocking conditions at placement in nursery sites. However, there is no evidence of that predictive models are currently routinely utilized in the swine industry to predict the nursery performance of individual groups of pigs based on their previous performance in the pre-weaning phase. One obstacle to building such predictive models is that health and management data as well as production data for the pre and post-weaning phases are collected with separate record-keeping programs and stored in unconnected databases.Thus, the objective of this study was to measure and compare the performance of 5 forecasting models for predicting nursery mortality using a master table containing data collected from a single swine production system for 1,831 groups of pigs. An automated model was built to integrate information for 48 variables included in the master table that was previously disconnected within the production system. After the model and parameter tuning process, the model with the best overall prediction performance was the Support Vector Machine (SVM) in terms of Root Mean Squared Error (RMSE=0.3929), Mean Absolute Error (MAE=0.2956), and coefficient of determination (R2=0.7761). This study demonstrated the ability to use an integrated master table containing whole-herd information until the placement event of nursery groups to predict their downstream nursery mortality through the initial 60 days of the post-weaning phase.
