The work aims to address the shortcomings of some plastic express bags, including high proportion of light elements, limited types of elements, and difficulties in effective differentiation by conventional X-ray fluorescence spectrometer. Infrared spectrometry and X-ray fluorescence spectrometer based on Johansson-type Doubly Curved Crystal bending technology were combined to analyze the composition and element content of 36 samples from 7 express delivery companies in different regions and batches. Firstly, infrared spectrometry was used to distinguish the sample matrix, followed by further subgrouping based on element content. Finally, the classification results were incorporated into gradient boosting decision tree model for validation of model robustness. The samples were divided into two groups by infrared spectrometry and further subdivided into five categories based on titanium and sulfur content. After optimization of the model parameters, the accuracy of the gradient boosting decision tree model reached 94.2%. This method can quickly classify and predict the types of samples based on differences in the matrix and fillers, filling the gap in conventional X-ray fluorescence spectrometer inspection, and providing a new investigation idea and technical means for tracing and classifying unknown plastic express bags.