Body Weight Prediction from Linear Measurements of Icelandic Foals: A Machine Learning Approach.
Authors: Satoła Alicja, Łuszczyński Jarosław, Petrych Weronika, Satoła Krzysztof
Journal: Animals : an open access journal from MDPI
Summary
# Editorial Summary Accurate bodyweight assessment in young stock is essential for tailoring nutrition programmes and detecting developmental concerns, yet many working practitioners lack access to weighbridges, particularly in breeding operations. Polish researchers applied machine learning algorithms to 312 biometric measurements collected from 24 Icelandic foals over their first 14 months, developing polynomial models that incorporated heart girth, body circumference, and cannon bone circumference to predict bodyweight with mean percentage errors of just 3.8–4.1%. Encouragingly, a simplified single-variable model using the square of heart girth multiplied by body circumference achieved comparable accuracy (up to 5% error), making practical field application viable for farriers, vets, and nutritionists without specialist equipment. These findings suggest that machine learning approaches can generate breed-specific, age-adjusted prediction equations superior to traditional linear regression formulas currently in use. For practitioners managing Icelandic or similar stock types, implementing such refined biometric models could substantially improve growth monitoring, feeding protocols, and early identification of developmental orthopaedic disease—particularly valuable where direct weighing is logistically challenging.
Read the full abstract on PubMed
Practical Takeaways
- •When weighbridges are unavailable, simple biometric formulas (heart girth and body circumference) can reliably estimate foal weight with <5% error, enabling accurate feeding and health monitoring decisions
- •The simplified single-trait model offers a practical field alternative requiring only two measurements, making it accessible for breeders without complex equipment
- •Regular body weight monitoring using these validated formulas supports appropriate nutrition planning and early detection of growth abnormalities in young horses
Key Findings
- •Polynomial machine learning model using heart girth, body circumference, and cannon bone circumference achieved 4.1% mean percentage error via cross-validation and 3.8% on holdout dataset
- •Simplified model using heart girth squared multiplied by body circumference achieved up to 5% mean percentage error with reduced complexity
- •Study included 312 measurements from 24 Icelandic foals (12 colts, 12 fillies) aged birth to 404 days
- •Machine learning methods can provide practical alternatives to weighbridges for estimating foal body weight using linear biometric measurements