The exploration geochemical method is one of the most effective methods to quickly delineate
regional prospective areas. Although this method can quickly delineate geochemical prospective areas, it often
ignores the spatial variability of geochemical backgrounds, potentially missing weak geochemical anomaly. In order
to identify multivariate geochemical anomaly in complex geological environments, the authors select a one-class
support vector machine (OCSVM) model for this study. The model allows for high-dimensional anomaly detection
without making any assumptions about data distribution. Taking the Amuhuiyilete region as an example, based on
the 1 ∶ 50 000 regional geological survey results, the data of 11 geochemical elements from stream sediments in the
study area were gridded using Surfer software. The gridded “true” data were generated based on the spatial loca
tions of known mining points in the study area. The spatial correlation between each geochemical element and the
known mining points was statistically analyzed, and elements with significant correlation to the known mining points
and concentrated elemental distribution were identified as prospecting indicator elements. In the study area, three
indicator elements were selected. The gridded data of these three indicator elements were used as input data for
OCSVM to conduct multivariate geochemical anomaly identification research. The models were optimized using both
the trial-and-test method and the artificial bee colony (ABC) optimization algorithm. The output results of both
models were obtained and combined with the “true” data, receiver operating characteristic (ROC) curves were
then plotted for the models optimized by the trial-and-test method and the ABC optimization algorithm, and corre
sponding area under the curve (AUC) values were calculated. The results show that the AUC value of the model
optimized by the trial-and-test method is 0. 879 6, while the AUC value of the model optimized by the ABC algo
rithm is 0. 897 8. At the same time, the proportion of anomalous grids identified by the two models is 27. 14% and
23. 65%, respectively. This indicates that, in the anomaly detection task, the model optimized by the ABC
algorithm performs slightly better than the model optimized by the trial-and-test method. The OCSVM optimized by
the ABC algorithm is more effective in identifying anomalous data points, and improving the overall model accuracy.