Near-infrared spectroscopy (NIRS), which leverages molecular absorption of infrared light, possesses strong “fingerprint” identification capabilities. It has become an indispensable tool in compound identification, molecular structure analysis, and quantitative analysis. In recent years, NIRS, combined with chemometrics, has shown tremendous potential for rapid, nondestructive detection in complex systems. In this study, near-infrared spectra of 48 yak meat samples with varying adulteration levels (400–2500?nm) were collected. After dividing the samples by concentration gradient, six multivariate calibration methods were compared: principal component regression (PCR), partial least squares (PLS), least absolute shrinkage and selection operator (LASSO), support vector regression (SVR), backpropagation artificial neural network (BP-ANN), and extreme learning machine (ELM). The PLS model was chosen as the base model. Furthermore, 14 spectral preprocessing methods were evaluated, including Savitzky–Golay smoothing, standard normal variate (SNV), multiplicative scatter correction (MSC), derivative spectra, and continuous wavelet transform (CWT). The CWT method was found to provide the best results. On this basis, seven variable selection algorithms, including uninformative variable elimination (UVE), Monte Carlo UVE (MCUVE), randomization test (RT), competitive adaptive reweighted sampling (CARS), whale optimization algorithm (WOA), gray wolf optimizer (GWO), and butterfly optimization algorithm (BOA), were applied. The WOA algorithm significantly improved model performance, selecting 11 and 21 key features for yak meat and beef content predictions, respectively. The final model achieved RMSEP values of 4.8607 (Rp?=?0.9157) for yak meat and 3.8066 (Rp?=?0.9486) for beef content. This study established the optimal modeling strategy for quantitative analysis of yak meat adulteration.