High prices of mineral resources and progresses in models of mineral deposits maintain a high activity in mining exploration. Many data integration techniques exist to assess the mining potential of a region, and it may be difficult for the practitioner to select the most appropriate. This paper compares two types of data-driven methods to estimate a probability map of mineralization occurrences from several sources of information: multivariate logistic regression, under the assumption that all indicators are independent, and weights of evidence, which assumes only conditional independence. A proposal to depart from conditional independence assumptions in Weights of Evidence is also presented, that allows a better management of data redundancy. All three methods are applied to gold potential mapping in a 3D geo-model of the Duparquet region located on the Porcupine Destor fault in the Abitibi subprovince, Canada.