Tilapia is one of the major farmed fish species in Taiwan, with an annual production about 60,000 mt. It is a common item in the daily diet. Therefore, this study used Nile tilapia (Oreochromis niloticus) as the model to determine fish freshness using near-infrared spectroscopy (NIR). A near-infrared spectrometer was used to scan the sectors of six individual sampling points on tilapia meat and skin to collect spectra. Volatile basic nitrogen (VBN) was also detected in six sampling points of tilapia fillets. Then, the NIR spectra and VBN values were used to train a machine learning model with MATLAB software to develop a real-time rapid detection method to determine fish freshness. Results showed that tilapias could be categorized into three groups: (I) with an average VBN of 13.91 ± 3.47 mg% when refrigerated at 4 ℃ for 20 hr; (II) with an average VBN of 16.65 ± 3.56 mg% when stored at 37 ℃ for 5 hr; and (III) with an average VBN of 23.76 ± 3.48 mg% when stored at 37 °C for 6 hr. The NIR spectra and VBN values were compared using MATLAB and a concatenating support vector machine. The highest rates of recognition of fish meat and skin reached by NIR and VBN models with good recognition capacity (above 85%) were 87.7% and 88.3%, respectively. This analysis provides a rapid and non-destructive method for detecting the freshness of fishery products, which can be used in the self-management of fishery product suppliers, supermarkets, and group meal suppliers.