Table 3.

Conceptual, strengths, limitations, comparison of machine learning (ML), and mathematical models (MM) used in livestock research.

Comparison	ML	MM	References
Approach	Artificial intelligence subsets enable systems to involve training algorithms, identify patterns, and make decisions with minimal human intervention, using large datasets to improve their performance over time.	They are used to understand the relationships between variables and predict outcomes based on established mathematical principles and theory.	(75, 76)
Data processing	ML algorithms require substantial data processing capabilities, particularly those for deep learning. The need for large volumes of data and the iterative nature of training often require significant computational resources.	These models process data using predefined equations and relationships, which are less data-dependent and require less processing power. The data requirements are often specific and may not need to be as large as those for ML models.	(76)
Complexity and adaptability	By retraining, ML algorithms can handle complex, high-dimensional data and adapt to new data, which may be less computationally intensive than redesigning a mathematical model.	While complex, they often require a deep understanding of the underlying processes to be accurately formulated.	(76)
Development process	The development process involves data preprocessing, feature selection, model selection, training, and validation.	The process involves defining the problem, formulating equations based on theoretical knowledge, solving these equations, and validating the model against known data.	(77)
Real-time processing	Real-time ML processing can be more challenging, especially with large models, as it may require substantial computational power to process data and make predictions promptly.	Since mathematical models are often based on simple calculations, they can be processed in real time, making them suitable for applications requiring immediate responses.	(78)
Limitations
	•ML models, especially deep learning models, can be considered “black boxes” because it's often difficult to understand how they arrive at their predictions. Compared to ML, MM models are generally more transparent and interpretable because their structure and variable relationships are explicitly defined.	•Scalability can be an issue if the model's complexity grows with the problem's size, leading to computational challenges.	(78–80)
	•MLs are heavily dependent on data. They require large amounts of data to train and improve their accuracy.	•Adapting them to new data or conditions may require a complete reevaluation and reformulation.
	Wearable sensors, imaging systems, and automated monitoring platforms produce high-resolution data streams that require substantial computing resources, including reliable internet access and high-performance hardware. For livestock systems in low- and middle-income areas, where technological infrastructure and qualified workers may be limited, these criteria pose serious challenges. The ongoing maintenance and updates of ML models also drive long-term operational costs.	•While data can be used to parameterise these models, MM are not solely reliant on data. They are based on theoretical understanding and can be developed with or without empirical data.