Abstract: The pitchblende/uraninite of different types of uranium deposits has different composition of rare earth elements, which can be used as an important index to distinguish the types of uranium deposits. Using the Python language-based classification model combined with principal component analysis (PCA) and support vector machines (SVM), the data of 216 groups of pitchblende/uraninite rare earth elements collected from six known uranium deposits worldwide were studied. With the 216 groups of data as the training set, the SVM classification model was constructed through data cleaning, feature scaling, PCA feature extraction, grid search and cross-validation for parameter optimization, and 24 groups of syn-metamorphic Hujiayu uraninite were intelligently identified. The test accuracy of the 14-dimensional training set optimal model to determine the type of Hujiayu uraninite using only rare earth elements is 0.4%,and the test accuracy of the optimal model of the 17-dimensional training set composed of rare earth elements, total rare earth elements, ratio of light and heavy rare earth elements,and europium anomalies is 75.0%, an improvement of 74.6% over the 14-dimensional training set. The model has a strong generalization ability. But through traditional rare earth element distribution curve and the w(ΣREE)-(LREE/HREE)_N diagram, the type of Hujiayu uraninite cannot be determined effectively. This study shows that the PCA-SVM algorithm can effectively determine the genetic type of uranium oxides by mining the data set with the addition of traditional rare earth discriminating indicators, and the effect is significantly better than the pure rare earth element data set and the traditional rare earth distribution curve, w(ΣREE)-(LREE/HREE)_N diagram.

Key words: uranium oxide, rare earth elements, traditional rare earth discriminant index, principal component analysis, support vector machine, classification