Proposing a short version of the Unesp-Botucatu pig acute pain scale using a novel application of machine learning technique.

Surgical castration of males is carried out on a large scale in the US swine industry and the pain resulting from this procedure can be assessed using the Unesp-Botucatu pig composite acute pain scale (UPAPS). We aim to propose a short version of UPAPS based on the behaviors best-ranked by a random forest algorithm. We used behavioral observations from databases of surgically castrated pre-weaned and weaned pigs. We trained a random forest algorithm using the pain-free (pre-castration) and painful (post-castration) conditions as target variable and the 17 UPAPS pain-altered behaviors as feature variables. We ranked the behaviors by their importance in diagnosing pain. The algorithm was refined using a backward step-up procedure, establishing the Short UPAPS. The predictive capacity of the original and short version of the UPAPS was estimated by the area under the curve (AUC). In refinement, the algorithm with the five best-ranked behaviors had the lowest complexity and predictive capacity equivalent to the algorithm with all behaviors. The AUC of Short UPAPS (89.62%) was statistically equivalent (p = 0.6828) to that of UPAPS (90.58%). In conclusion, the proposed Short UPAPS might facilitate the implementation of a standard operating procedure to monitor and diagnose acute pain post-castration in large-scale systems.