Body Weight Prediction from Linear Measurements of Icelandic Foals: A Machine Learning Approach

SIMPLE SUMMARY: Knowing the body weight of a growing horse is important both for horse breeders and veterinarians because this information helps to identify abnormalities of the growing process, determine adequate feeding rations or choose an appropriate drug treatment regimen. It is not always possible to measure accurately a horse’s body weight using special scales, and a visual assessment, which is the easiest method for finding out a horse’s body weight, produces heavily biased results. Simple formulas are being sought to allow making accurate estimates of body weight in horses based on their body measurements. This study relates the estimation of body weight in Icelandic foals with the use of models relying on machine learning methods. Based on their evaluation, two of the models are recommended for use in practical applications. ABSTRACT: Knowledge of the body weight of horses permits breeders to provide appropriate feeding and care regimen and allows veterinarians to monitor the animals’ health. It is not always possible to perform an accurate measurement of the body weight of horses using horse weighbridges, and therefore, new body weight formulas based on biometric measurements are required. The objective of this study is to develop and validate models for estimating body weight in Icelandic foals using machine learning methods. The study was conducted using 312 data records of body measurements on 24 Icelandic foals (12 colts and 12 fillies) from birth to 404 days of age. The best performing model was the polynomial model that included features such as heart girth, body circumference and cannon bone circumference. The mean percentage error for this model was 4.1% based on cross-validation and 3.8% for a holdout dataset. The body weight of Icelandic foals can also be estimated using a less complex model taking a single trait defined as the square of heart girth multiplied by body circumference. The mean percentage error for this model was up to 5% both for the training and the holdout datasets. The study results suggest that machine learning methods can be considered a useful tool for designing models for the estimation of body weight in horses.