Climate and non-genetic effects on goat milk quality and casein prediction using machine learning

Abstract

Milk quality is becoming more and more crucial for farmers with the increasing use of component-based pricing systems in the sector. This study aimed to investigate the effects of climate-related and non-genetic factors on goat milk quality on the island of Northern Cyprus and to predict casein content using machine learning techniques. An analysis was performed on 21,695 goat milk samples collected between 2012 and 2018. Somatic cell count (SCC), standard plate count (SPC), and key milk components such as protein, casein, lactose, and fat were evaluated for different seasons, regions, farm sizes, and raw milk types. Five machine learning models were tested to predict casein content. SCC was significantly affected by year, season, farm size, and milk storage method, with higher values observed in larger farm types and non-chilled milk. On the other hand, the temperature-humidity index (THI) had a negative effect on milk composition, particularly reducing protein, casein, and lactose levels. Among the five machine learning models tested to predict casein content, the Random Forest Regressor achieved the highest accuracy rate (R² = 0.988, RMSE = 0.041). Climate-related factors and farm management practices significantly affected goat milk quality. Combining the climate-adapted management strategies with data-driven forecasting tools can help maintain milk quality under variable environmental conditions.