Abstract: This study aims to develop a methodology for implicit 3D geological modeling under the constraint of limited exploration data availability. Taking Lazigou gold deposit as an example, we generate virtual grooving by distance weighting method based on the original grooving sampling data, and use the original grooving data to train and evaluate three machine learning models, namely K-nearest neighbor, random forest and gradient elevator, and select the random forest model with the best prediction performance to discriminate and predict the lithology of virtual grooving. Virtual encryption of groove data is realized by machine learning method, which provides a large number of sample data for implicit 3D modeling. On this basis, the orebody model and Au element grade numerical model of Lazigou gold mine were constructed in implicit modeling software using the original groove and virtual groove data. Five prospecting targets have been delimited, which have been proved reliable by engineering. The implicit 3D modeling based on machine learning can make full use of the known data to predict the unknown region and provide sufficient samples for the implicit 3D modeling, which is conducive to the construction of a higher precision geological model under the existing exploration engineering conditions, and then provide a basis for the deep edge prospecting prediction.