326 - Nutritional Grouping and Machine-learning Techniques: Towards a Feed Efficiency Improvement in Beef Cattle Production

Abstract: Nutritional grouping (NG) strategies have been proposed mainly in dairy cattle to improve economic efficiency (30-60$/cow per year) of dairy farmers. In dairy cattle, NG is made according to net energy (NE) requirements estimated from some animal variables (i.e., milk yield and dairy merit). However, NG has not been evaluated for beef cattle, likely because data are not readily available and existing protocols have been fixed for decades by the beef industry, such as allotting by body weight (BW): growing diet for 180 to 400-kg animals, and finishing diet for 400 to 550-kg animals. Therefore, we investigated an unsupervised machine learning (ML) technique in R to sort animals at the arrival according to their DM intake (DMI), NE and metabolizable protein (MP) requirements, estimated through INRA from BW and average daily gain. We utilized data from 77 growing Friesian steers allotted conventionally by BW (139±21.6 kg) in 4 pens, weighed every 3 weeks (n=9 weighings) during 190 days, and fed a high-energy diet (1.59 Mcal NE/kg DM and 89 g MP/kg DM). Results identified that estimated performances were significantly (P < 0.002) different between 3 groups made from ML and between 4 pens made from BW; however, ML groups presented a minor within-group variability of NE and MP requirements (-25%CV) than pens. We also regressed energy and protein intake minus their respective requirements by the duration of the experiment observing a better fitting (R2= 0.93 vs 0.80) when allotting animals by groups than by pens, which implicates better feed efficiency if diets were formulated according to requirements established by ML groups. We demonstrated that ML techniques could help the decision-making of beef farmers to better sort animals according to their requirements and not only by BW. The next step is to formulate rations according to animal requirements to improve efficiency.