Use of pattern recognition techniques for early detection of morbidity in receiving feedlot cattle 1

Abstract

Two groups of cattle were used to develop (model data set: 384 heifers, 228 ± 22.7 kg BW, monitored over a 225-d feeding period) and to validate (naïve data set: 384 heifers, 322 ± 34.7 kg BW, monitored over a 142-d feeding period) the use of feeding behavior pattern recognition techniques to predict morbidity in newly arrived feedlot cattle. In the model data set, cattle were defined as morbid (MO) if they were removed from their pen to be treated due to visual observation of clinical signs of bovine respiratory disease and healthy (HL) if they remained within their pen and lacked lung lesions at slaughter. Individual feeding behavior parameters collected with a GrowSafe automated feeding behavior monitoring system were reduced via principal component analysis to 5 components that captured 99% of the variability in the data set. Combinations of clustering and cluster classification strategies applied to those components, along with pattern recognition techniques over different time windows, produced a total of 105 models from which precision, negative predictive value, sensitivity, specificity, and accuracy were calculated by comparing its predictions with the actual health status of individual cattle as determined by visual assessment. When the models with the best specificity (models 79 and 87), sensitivity (models 33 and 66), and accuracy (models 3 and 14) in the model data set were used in a naïve data set, models 79 and 87 were not able to predict any MO heifers (0%), with all animals (100%) being predicted as HL. Model 33 predicted 58.3% of the HL and 66.7% of the MO heifers, with MO heifers identified 3.1 ± 1.64 d earlier than by visual observation. Model 66 predicted 50.0% of the HL and 75.0% of the MO heifers, with MO heifers predicted 3.1 ± 1.76 d earlier than by visual observation. Model 3 predicted 100% of the HL and 50.0% of the MO cattle, with MO cattle predicted 1 d earlier than by visual observation. Model 14 predicted 83.3% of the HL and 58.3% of the MO cattle, with MO cattle detected 2.4 ± 1.99 d earlier than visual observation. The application of pattern recognition algorithms to feeding behavior has potential value in identifying MO cattle in advance of overt physical signs of morbidity. Work on an integrated system that would automatically process data collected from automated feed bunk monitoring systems is still required, however, for this method to have value to the commercial feedlot industry as a practical means of identifying MO cattle in real time.

Invited review: Big Data in precision dairy farming

“…The paper of Moya et al . (2015) is the only paper that addresses three different machine-learning techniques, reporting on supervised classification, unsupervised clustering and dimensionality reduction. The papers from Martinez-Ortiz et al .…”

Evaluation of statistical process control procedures to monitor feeding behavior patterns and detect onset of bovine respiratory disease in growing bulls

“…Lukas et al (2008) reported preclinical reductions in DMI in dairy cows with mastitis and reductions in water intake associated with the febrile response. Using pattern recognition techniques, Moya et al (2015) reported that morbid cattle exhibit distinctive deviations in feeding behavior patterns prior to displaying overt clinical signs of BRD and can be differentiated from the feeding patterns of healthy cattle. Based on discrete survival time analysis of DMI and feeding behavior data, Wolfger et al (2015) reported that increases in DMI per meal, meal frequency and intermeal interval were associated with a decreased hazard for developing BRD up to 7 d prior to observed clinical signs of disease.…”

“…Few studies have been conducted that have monitored changes in behavioral or physiological patterns of individual animals to predict the onset of disease. Other attempts at early identification have used hazard analyses (Wolfger et al, 2015), mean comparison (Sowell et al, 1999), logistic regression (Schaefer et al, 2007), and cluster analyses (Moya et al, 2015). There are also studies and commercial products available that have published results without clearly defining the method or described the algorithm that was used to predict BRD.…”

“…Most of the studies that have monitored animal health have all used high-risk type cattle; Quimby et al (2001) monitored cattle that had an overall pull rate of 67% compared to 13% in the current study. Other morbidity rates range from 77% (Wolfger et al, 2015) to 29 % (Moya et al, 2015), which are much higher than normal (mean morbidity rate = 14%, Irsik et al, 2006). It is unknown whether using high-risk cattle to develop a predictive model would affect the ability of the model to select morbid animals in a normal population.…”

A glimpse of the future in animal nutrition science. 2. Current and future solutions

“…Detection of SPC-model of BRD based on feed bunk attendance and head-down duration occurred 2.7 and 3.0 days prior to observed BRD diagnosis, with model accuracies of 87 and 89%, respectively. More recently, Moya et al (2015) used nonlinear data-mining analysis of feeding behavior data to develop pattern-recognition-based algorithms to predict morbidity events in beef cattle. When validated against a naive group of calves, they were able to correctly predict the health status of high-risk calves with a model accuracy of 79%.…”