Abstract: Zircon is widespread and compositionally stable in intermediate–acid magmatic rocks and is resistant to later hydrothermal activities. Therefore, its composition can more accurately record information about mineralizing magmas. Among them, zircon features (such as Ce⁴⁺/Ce³⁺, Ce/Ce^*, Eu/Eu^*, and Ce/Nd) have been widely used in evaluating the mineralization potential of granitoids, because they have been found to reflect ore−forming information, such as magmatic oxygen fugacity and water content. However, further studies have revealed that the universality of these geochemical indicators has been questioned. In addition, the proposed methods for discriminating mineralization capacity are all based on the current “limited understanding” of mineralized rocks, and considering the complexity of the mineralization process, much geochemical information reflecting the capacity of magmatic mineralization may not have been revealed yet. Therefore, in the paper, taking the Qimantagh mineralized zone of the East Kunlun as an example, and with the help of one of the most widely used machine learning algorithms today (Support Vector Machine), the authors trained machine learning on zircon data from porphyry skarn Cu−Fe−Pb−Zn mineralized rock bodies in the region and zircon data from non−mineralized rock bodies around the world, and the aim is to excavate zircon trace element signatures that reflect magmatic mineralization capacity, so as to construct a new discriminative schema for granite mineralization potential. The results of the model training show that among 21 common zircon trace element features, five element features, Gd, Dy, Yb, Y and Tm are the most important for identifying the magmatic mineralization ability; based on this, 10 binary discriminant diagrams are established in this paper, and their accuracy rates in identifying mineralized and non−mineralized rock bodies are close to 1. The present study show that the use of machine learning methods and geological big data can be used to explore the potential of granite mineralization which is difficult to study with traditional research methods. The study demonstrates that machine learning methods and geological big data can be used to mine new geochemical indicators and diagrams that are difficult to discover by traditional research methods, which is of great significance to deeply understand the genesis of mineral deposits and guide the prospecting and exploration of minerals.