<p>Reliable seed viability assessment is critical for the ex situ conservation of forest genetic resources, yet traditional methods are destructive, labor-intensive, and time-consuming, highlighting the need for non-destructive alternatives. This study investigates the feasibility of using Fourier transform near-infrared (FT-NIR) spectroscopy combined with machine learning techniques for non-destructive viability assessment of <i>Pinus densiflora</i> seeds stored for 20 years at 4℃. FT-NIR spectra were collected from 600 seeds, followed by germination tests to determine seed viability. Various spectral preprocessing methods, including scatter correction and Savitzky–Golay derivatives, were compared to optimize model performance. Feature selection using forward sequential selection effectively reduced dimensionality while maintaining classification accuracy. Among 14 tested algorithms, ensemble methods such as Random Forest, Extra Trees, and Gradient Boosting classifiers consistently achieved high performance, with the best classification accuracy reaching 83.89% on the test set. This study demonstrates the applicability of FT-NIR spectroscopy and machine learning for viability assessment in seeds that have undergone gradual aging under cold storage conditions. These findings may serve as a practical foundation for developing non-destructive monitoring tools in real-world seed bank management.</p>