This study employed differential Raman spectroscopy to analyze 60 drug packaging carton samples, aiming to quantify and qualify the chemical fillers present. A novel method for rapid, non-destructive evaluation of diverse drug packaging materials was developed. The original spectrum is pre-treated by the standard deviation standardization method (Z-Score) and the sample is divided into five categories by the condensed hierarchical clustering algorithm. Classification was performed using a discriminant analysis model, which achieved accuracy rates of 96.7% and 78.3% for different models. Additionally, an artificial neural network model was trained and tested, yielding accuracy rates of 95.35% and 82.35%, respectively. To assess the predictive significance of characteristic wavelengths, a random forest model was also constructed using the same datasets, achieving higher relevance with accuracy rates of 90.8% and 85.0%. The findings demonstrate the non-destructive testing capabilities of differential Raman spectroscopy on various types of drug packaging cartons, as well as the efficiency and speed of the general prediction model, which establishes the groundwork for quick examination and evaluation of drug packaging cartons.