The work aims to establish a color prediction model for water-based inks by utilizing near-infrared spectroscopy and spectral fusion strategies combined with chemometrics methods, to achieve accurate color prediction of ink prints. The near-infrared spectral reflectance and absorbance data of inks with different alcohol contents and color ink contents were collected and the corresponding Lab values of prints were measured. Then, a partial least squares (PLS) model for a single spectrum after different preprocessing and a PLS model based on data layer fusion and feature layer fusion were established. Finally, the determination coefficient of the prediction set with the root mean square error of the prediction set (RMSEP) and the chromatic aberration were compared to evaluate the predictive performance of the model. For single spectral modeling, the accuracy of the model based on reflectance was higher than that of the model based on absorption. The lack of stability in data layer fusion improved the prediction of L and b values, while the prediction of a value remains almost unchanged. The modeling effect of feature layer fusion was significantly better than that of single spectrum and data layer fusion, with prediction determination coefficients for Lab reaching 0.996 1, 0.993 9, and 0.997 4, respectively. The RMSEPs were 0.142 1, 0.212 6, and 0.207 2, respectively. The maximum color difference between the predicted value and the true value was 0.678 3. Spectral feature fusion technology can improve the accuracy of ink color prediction and accurately predict ink color changes.