Discriminant Analysis.pptx

Discriminant Analysis • CASE STUDY: Use of Individual Feeding Behavior Patterns to Classify Beef Steers into Overall Finishing Performance and Carcass Characteristic Categories

About case • Three hundred seventy beef steers were used to evaluate the possibility of using individual animal feeding behavior patterns early in the finishing period to classify overall finishing performance, carcass quality, or both. • Individual animal performance records for the overall finishing period were sorted into classification groups including: 1) ADG quartiles, 2) hot carcass weight (HCW) quartiles, 3) quality grades (QG) groups of Choice + Prime or Select + Standard, 4) yield grade (YG) groups of YG 1 + YG 2 or YG 3 + YG 4, and 5) liver abscess groups of no abscesses or abscessed livers at the time of slaughter. • Animals correctly classified into performance classification groups ranged from 1.4 to 60% for ADG quartiles; 23.2 to 62.9% for HCW quartiles; 63.5 and 58.3% for Select + Standard and Choice + Prime, respectively; and 62.1 and 55.7% for YG 1 +YG 2 and YG 3 + YG 4, respectively.

• Shifts in the beef industry have resulted in an increased emphasis on quality-based marketing. • Pricing is often based on the producer’s ability to correctly identify groups of animals that are most likely to yield the desired carcass qualities. • Results of previous research (Hicks et al., 1989) suggest feeding behavior patterns are highly repeatable366 Parsons et al. and that frequency of feeding is positively related to growth of feedlot cattle; however, collecting and quantifying direct relationships among animal behavior, health, and performance have been difficult.

• Whitley (2000) suggested that individual feeding behavior data collected with the GrowSafe system might be useful to predict cattle performance and carcass measurements. • Discriminant analysis is a multivariate statistical method that is useful for studying the extent to which different populations are related and de-fining a predictive function for classification of data. • Our objective was to evaluate the ability of individual animal feeding behavior data obtained using the GrowSafe system to classify feedlot cattle into overall performance or carcass outcome groups by discriminant analysis.

Materials and Methods • Three hundred seventy medium- and large framed beef steers were used in a feeding behavior trial conducted at Cactus Research. • . At processing, each steer was identified with two uniquely numbered ear tags, individually weighed, implanted. • The steers were allotted randomly to four consecutive pens, and the number of steers assigned to each pen was adjusted to provide approximately 22.9 cm of feed bunk space and 13.9 sq.m of pen space per head. • The steers were fed twice.

Feeding Behavior System • GrowSafe system recorded individual animal identification. The electronic monitoring system consisted of four main components: 1) individually numbered transponder ear tags. 2) a neoprene mat that contained an antenna to receive signals from the ear tags 3) a reader panel 4) a desktop computer

Experimental Design • Individual steers were the experimental unit for collection and analysis of feeding behavior data. • As described earlier, the steers included in this study were fed in four different pens. • To adjust forpossible pen effects, the feeding behavior data for individual animals were divided by a pen-adjustment factor. • Pen-adjustment factors were calculated by dividing the pen average for each feeding behavior variable by the overall average for the variable. • Two steers died, eight were eliminated for chronic health problems, and 25 were eliminated because of missing feeding behavior data

Measurements • The five measurements used to define an individual feeding behavior pattern were as follows: • Head-down duration (HDD), i.e., the number of times the animal was recorded by a mat reader multiplied by 6.3 s; • In-to-out duration (ITOD), i.e., total duration of the feeding event from initial presence at the bunk to the final reading not broken by a ≥300-s absence; • Feeding frequency (FREQ), i.e., the number of visits to the feed bunk per day; • Ratio 1 (INT1), i.e., HDD ÷ FREQ; • Ratio 2 (INT2), i.e., HDD ÷ ITOD.

Results and Discussion These statements led to two assumptions that we used in the interpretation and discussion of these results: 1) to be applicable to the beef cattle industry, classification of overall performance or carcass characteristics using feeding behavior pattern must be accurate; and 2) to eliminate excessive handling, classification must be possible before cattle are reimplanted. • Commercial feedlot pens in the Great Plains region of the U.S. typically contain 50 to 200 animal. • The ITOD could be used to classify the lowest 25% of animals defined by ADG with an accuracy of 60% • Accuracy of the linear discriminant function for the other three comparison periods (data not shown) averaged 34.9, 27.4, 19.8, and 36.4% for the lowest to highest ADG quartiles, respectively. • The linear discriminant function, based solely on initial BW, correctly classified 62.9, 23.2, 29.0, and 62.9% of steers into the lightest to heaviest HCW quartiles, respectively.

Results and Discussion • These preliminary results were somewhat hampered by equipment failures. • Nonetheless, based on these preliminary findings, accurately classifying feedlot cattle into overall performance and carcass characteristic outcome groups was not possible using a combination of the feeding behavior measurements and initial BW. • Feeding behavior patterns measured under different conditions and during different seasons, in addition to potential feeding behavior patterns not considered in this study, need to be evaluated in future research