Traceability of apple origin and prediction of apple SSC is of great practical significance, and the purpose of origin discrimination and SSC prediction is achieved by modeling. To overcome the limitations of a single model, the overall prediction performance is improved by combining the prediction results of multiple models. Near-infrared spectroscopy (NIRS) detection technology combined with a multi-model decision fusion strategy is utilized for traceability identification of apple origin and prediction of apple SSC to verify the feasibility of the theoretical method.The spectra of apple samples were collected using a handheld near-infrared detector, and apple origin discrimination models were established using the sample spectra in combination with the random forest (RF) method, the partial least squares discriminant analysis (PLS-DA) method, and the support vector machine (SVM) method. The predictions from the three discrimination models are then used in a voting system decision fusion method to generate new discriminant results. Actual values of SSC were collected for all apple samples, and SSC prediction models were developed using the sample spectra and actual values of SSC combined with the random forest (RF) method, the partial least squares regression (PLSR) method, and the support vector regression (SVR) method. Using the outputs of the three regression models, the new SSC prediction is output through the weighting method decision fusion strategy. When the voting decision-making method was not used, the discrimination modeling using the RF method was the most effective among the three qualitative modeling methods, with a prediction accuracy of 88.71%. The worst prediction was made using the SVM method, with a prediction accuracy of 77.43%. After using the voting decision method, the accuracy of apple origin identification reached 93.42%, and its prediction precision and recall also reached a double high, both above 85%. All three quantitative modeling methods gave good results in predicting apple SSC without using the weighted decision fusion method. All three methods predicted coefficients of determination around 0.87 and root mean square errors of prediction (RMSEP) around 0.78. The prediction of the SSC level was improved after using the weighted decision fusion method. The prediction coefficient of determination was 0.91, and the RMSEP was 0.66. The feasibility of the proposed method was confirmed by using the multi-model decision fusion method in the identification of apple origin and the prediction of apple SSC to improve the accuracy of apple origin discrimination and the precision of the prediction of apple SSC. Meanwhile, the handheld NIR detector combined with the multi-model decision fusion method provides a new high-precision prediction approach for on-site non-destructive testing analysis.