Dealing with Errors in Data from Electronic Swine Feeders

Abstract

<p>Data from electronic swine feeders contain errors that must be identified, edited, and corrected. The objectives of this study were to develop comprehensive criteria to identify errors in feed intake data from FIRE electronic feeders and to compare the ability of five editing methods to accurately estimate daily feed intake (DFI) and average daily feed intake (ADFI). Data from FIRE feeders on 591 pigs from the National Pork Board’s Maternal Line Genetic Evaluation Program were used. Errors in each visit were identified using 16 criteria. To create an error-free data set as a basis for comparison, data from 124 pigs with few errors were selected and visits with errors were replaced by error-free visits from the same pig. Resulting DFI and ADFI were assumed to be the true trait values. Error visits were then introduced, representative of field data. Data were edited using five methods (EM1-5). For EM1, a DFI record was deleted if DFI < 1000 g or >4500 g. For EM2-5, the 16 criteria were used to identify errors in each visit. For EM2 and 3, all DFI records with >1 and >2 error visits were deleted. For EM4-5, DFI was obtained by summing feed intake over error-free visits. For EM5, DFI records were then adjusted for the effects of presence of error visits on unadjusted DFI, which were estimated from a linear model analysis. For EM1-4, missing DFI records were replaced by linear regression estimates of DFI on test day for each pig. DFI and ADFI from the edited data sets were correlated to true values. Correlations were high (.90 to .99) for both traits for all editing methods except EM1. EM5 had the highest correlation for DFI (.99). EM2 and EM5 had the highest correlations for ADFI (.98 to .99). EM1 had the lowest correlations for both traits (.82 to .93). Results indicate that editing methods affect the accuracy of data from electronic feeders. EM5 is recommended for maximum accuracy for DFI and EM2 is recommended for ADFI for maximum accuracy and ease of implementation.</p>