A recent study demonstrated how to use mid-infrared (MIR) spectral data to predict body condition score (BCS) changes in dairy cows.
A multidisciplinary study demonstrated that herd management of cows can be improved by using milk samples to predict the body condition score (BCS) in cows across national boundaries. This study investigated whether prediction models developed using mid-infrared (MIR) data from one country could accurately forecast BCS changes in dairy herds located in another (1). However, the findings, which were published in the Journal of Dairy Science, suggest that cross-border data integration offers limited benefits and may even reduce predictive accuracy (1).
BCS evaluates the fatness and thinness of dairy cows on a five-point scale. A score of 1 indicates that the cow is thin and malnourished, whereas a 5 indicates that a cow is severely overweight (2). The BCS score is designed to help farmers maintain consistent milk production and ensure it is of high quality. The BCS also helps dictate dairy heifer feeding management (2). Even small daily changes in this score can have a significant impact on a cow's health and milk production. As a result, maintaining an optimal BCS helps ensure that cows are neither overfed nor underfed, reducing risks related to metabolic disorders, fertility issues, and overall milk yield (1,2).
In this study, routine milk samples underwent spectroscopic analysis in the MIR region, a technique widely used to estimate milk macro-constituents like fat, protein, lactose, and urea content (1). The researchers sought to use MIR data to predict daily changes in BCS (ΔBCS), exploring the potential for transferring prediction models across countries—specifically between Canada and Ireland.
The study involved an extensive data set of 347,254 BCS records from 80,400 cows in Canada and 73,193 records from 6,572 cows in Ireland. Researchers applied two prediction methods—partial least squares regression (PLSR) and neural networks (NN)—to evaluate the ability of MIR spectra to predict daily ΔBCS. They tested two key scenarios. The first was combining Canadian and Irish data to develop a prediction model and applying it to each country separately (1). The second was using the Canadian and Irish data individually to predict daily ΔBCS in each country (1). Additionally, the research assessed the impact of data pretreatment (for example, applying the first derivative of the spectrum) and standardization of ΔBCS across countries on prediction accuracy (1).
The researchers achieved mixed results on cross-country predictive ability. The prediction models developed using MIR data from one country did not perform well when applied to another. When the Canadian MIR data were used to predict Irish ΔBCS, and vice versa, the correlation between actual and predicted BCS scores dropped significantly (1). The correlation between actual and predicted ΔBCS in the same country, however, remained high, ranging from 0.92 to 0.94 for the Canadian data and 0.85 to 0.87 for the Irish data (1).
The study also found that the combination of Canadian and Irish data for model development did not enhance predictive ability for either country. The correlations between actual and predicted ΔBCS in both countries were lower (≥0.90 for Canada and ≥0.80 for Ireland) compared to models developed using data from just one country (1).
As a result, the research highlighted both the promise and limitations of using advanced data analytics in dairy herd management. Although the study demonstrated the utility of using MIR spectral data to predict BCS within a country, the transferability of these prediction models across borders remains a challenge (1). This is significant for international dairy operations and researchers aiming to share and apply data-driven solutions globally.
