Objective To establish a model for the rapid quantification of sorghum starch based on near-infrared spectroscopy and machine learning algorithms.Methods Brewing sorghum was used as the raw material. Near-infrared diffuse reflectance spectra were collected using a near-infrared spectrometer. The Isolation Forest algorithm was employed to remove outlier samples and optimize the dataset. Subsequently, multiple preprocessing methods and regression models were compared using various modeling software to analyze and establish a quantitative model for sorghum starch.Results After preprocessing with the Isolation Forest algorithm, the partial least squares regression (PLSR) model exhibited the best performance, with a coefficient of determination for the calibration set (Rc2) of 0.993 and a root mean square error of cross-validation (RMSECV) of 3.401.Conclusion The model is simple, rapid, and accurate, and is suitable for the quantitative analysis of starch content in sorghum.