A predictive model for detecting the addition of milk to yak milk was established by combining mid-infrared spectroscopy （MIRS） with machine learning algorithms to supervise and regulate the production and sale of yak milk， further quantitatively predict the proportion of adulteration in yak milk and provide new technology of rapid detection. 76 samples of pure yak milk， 76 samples of yak milk adulterated with 10% milk， 76 samples of yak milk adulterated with 25% milk， and 76 samples of yak milk adulterated with 50% milk were used to establish binary qualitative models for distinguishing pure yak milk from yak milk adulterated with milk， and quantitative regression models for predicting the proportion of yak milk adulterated with milk with five spectral preprocessing algorithms， six qualitative and twelve quantitative machine learning algorithms. The results showed that the predictive model for identifying pure yak milk and yak milk adulterated with milk based on support vector machine modeling algorithm and the spectrum without preprocessing had the best performance. The validation set AUC， the accuracy， the sensitivity， and the specificity of the model was 0.95， 0.84， 0.93， and 0.87， which can be used for the identification of pure milk and adulterated milk. The optimal quantitative model for predicting the proportion of milk in yak milk was established using Bayesian regularized neural network modeling algorithm and first-order derivative spectral preprocessing algorithm. The model had R_P²=0.88， RMSEV=6.57%， and RPD=2.89%. It is indicated that the combination of mid-infrared spectroscopy and machine learning algorithms can effectively identify yak milk adulterated with milk and detect the proportion of adulteration.