The traditional plastic sorting process primarily relies on manual operations, which are inefficient, pose safety risks, and result in suboptimal separation efficiency for mixed waste plastics. Near-infrared (NIR) spectroscopy, with its rapid and non-destructive analytical capabilities, presents a promising alternative. However, the analysis of NIR spectra is often complicated by overlapping peaks and complex data patterns, limiting its direct applicability. This study establishes a comprehensive machine learning-based NIR spectroscopy model to distinguish polypropylene (PP) at different aging stages. A dataset of NIR spectra was collected from PP samples subjected to seven simulated aging stages, followed by the construction of a classification model to analyze these spectral variations. The aging of PP was confirmed using Fourier-transform infrared spectroscopy (FTIR). Mechanical property analysis, including tensile strength and elongation at break, revealed a gradual decline with prolonged aging. After 40 days of accelerated aging, the elongation at the break of PP dropped to approximately 30%, retaining only about one-sixth of its original mechanical performance. Furthermore, various spectral preprocessing methods were evaluated to identify the most effective technique. The combination of the second derivative method with a linear -SVC achieved a classification accuracy of 99% and a precision of 100%. This study demonstrates the feasibility of the accurate identification of PP at different aging stages, thereby enhancing the quality and efficiency of recycled plastics and promoting automated, precise, and sustainable recycling processes.