In the current environment of rapid development of “big data”, the basic idea of integrating and exploring multiple types of information represented by “multi source heterogeneous” geoscience big data is consistent with the idea of “big data”. Based on the decision tree prediction model, this paper used the random forest algorithm to upgrade five mineral models. First, the machine learning mineral resource prediction was discussed, then the geological model was established using three-dimensional technology, then the three-dimensional quantitative prediction was carried out, and finally the relevant experimental analysis was carried out. The experimental results show that when using the random forest algorithm, the model accuracy can reach up to 95.8%. It was found that the random forest model had higher prediction accuracy and shorter prediction time. The research in this article was expected to provide new ideas and methods for predicting and evaluating the spatial distribution patterns of mineral resources in the future.