Statistical Analysis of Mid-Infrared Spectroscopy Data to support Dairy Cow Management

Mid-infrared (MIR) spectroscopy is a methodology that uses the absorption pattern of light in the MIR region of the electromagnetic spectrum to indirectly estimate the concentration of constituents in a biological sample. Mid-infrared spectroscopy is a low-cost, rapid and non-disruptive technique used to analyse milk composition in practically every laboratory worldwide. The objective of this thesis was to estimate both milk quality traits and cow energy status as well as to discriminate cow milk based on the diet fed by applying different statistical techniques to milk MIR data. The data used originated from a series of research studies over several years. Different statistical techniques (i.e., partial least squares regression, neural networks, ridge regression, LASSO, elastic net, random forest, boosting random forest, ensemble method, partial least squares discriminant analysis, linear discriminant analysis, support vector machines) were investigated. The correlation between the gold standard and predicted milk quality traits investigated ranged from 0.28 (i.e., casein micelle size) to 0.80 (i.e., milk pH), while the correlation between actual and predicted body condition score change (ΔBCS) was 0.87 in early lactation cows. The milk MIR spectrum was able to differentiate between cows fed grazed pasture or TMR with an accuracy of 96.8%. Accurately estimating milk quality, ΔBCS and diet from a milk MIR spectrum can be a useful decision support tool for dairy farmers and dairy processors. Milk quality traits estimated from MIR can be useful for dairy processors to segregate incoming milk for different production lines. Moreover, dairy processors can label their products as “Grass-Fed” with more confidence if this is substantiated from the milk MIR spectra. Cow ΔBCS estimated from the milk MIR spectrum can be used to support producer decisions at the individual cow level, particularly to identify cows losing more body condition than the mean of the herd at an early stage. Therefore, statistical analysis of MIR spectroscopy data is a useful tool to support dairy cow management.