Application of Supervised Machine Learning Techniques and Digital Image Analysis for Predicting Live Weight in...

Area of Science:

• Poultry Science
• Machine Learning
• Computer Vision

Background:

• Accurate live weight estimation is crucial for efficient poultry production and precision farming.
• Traditional methods of weighing can be labor-intensive and stressful for birds.

Purpose of the Study:

• To evaluate supervised machine learning (ML) algorithms and digital image analysis for non-invasive live weight prediction in Anadolu-T broilers.
• To assess the accuracy of ML models using morphometric traits and image-based data.

Main Methods:

• Collected 4200 records from 100 broilers (50 male, 50 female) over 42 days, measuring back length, width, and live weight.
• Trained and validated five ML algorithms: Random Forest (RF), k-Nearest Neighbors (KNN), Support Vector Regression (SVR), Extreme Gradient Boosting (XGB), and Multiple Linear Regression (MLR).
• Developed image-based models correlating body surface pixel area with live weight.

Main Results:

• KNN algorithm demonstrated the highest accuracy (R² = 0.982) among ML models, with RF and XGB also showing reliable predictions.
• Image-based models achieved high accuracy (R² = 0.989), indicating projected surface area effectively represents growth.
• Both ML and image analysis provided accurate, non-invasive live weight estimations.

Conclusions:

• Integrating ML algorithms with digital imaging offers a practical, cost-effective solution for real-time broiler weight estimation.
• This approach advances precision poultry farming through automated, data-driven growth monitoring.
• Non-invasive weight prediction using morphometric data and image analysis is feasible and accurate.