Near-infrared spectroscopy assisted by random forest for predicting the physicochemical indicators of yak milk powder

High-efficiency and cost-effective detection of physicochemical indicators is essential for the quality control of yak milk powder. Herein, a rapid and simultaneous detection method based on miniaturized near-infrared (NIR) spectroscopy and chemometrics for four physicochemical indicators (protein, fat, and moisture contents as well as acidity) of yak milk powder was developed. By comparing partial least squares combined with support vector regression (PLS-SVR), ridge regression (RR), and random forest (RF), the optimal prediction models were identified. The results indicated that the combination of RF and NIR spectroscopy achieved excellent performance in predicting the four indicators, with correlation coefficients of 0.9846, 0.9642, and 0.9915 for the protein, fat, and moisture contents, respectively, and 0.9819 for acidity. This method enables rapid and accurate prediction of yak milk powder quality, providing a reliable tool for production quality control. Future work should explore its scalability and integration into real-time monitoring systems.