In this paper, we introduce a hybrid approach for analyzing metabolomic data about the so-called diabetes of type 2. The identi-cation of biomarkers which are witness of the disease is very important and can be guided by data mining methods. The data to be analyzed are massive and complex and are organized around a small set of individuals and a large set of variables (attributes). In this study, we based our experiments on a combination of ecient numerical supervised methods , namely Support Vector Machines (SVM), Random Forests (RF), and ANOVA, and a symbolic non supervised method, namely Formal Concept Analysis (FCA). The data mining strategy is based on ten spe-cic classication processes which are organized around three main operations , ltering, feature selection, and post-processing. The numerical methods are mainly used in ltering and feature selection while FCA is mainly used for visualization and interpretation purposes. The rst results are encouraging and show that the present strategy is well-adapted to the mining of such complex biological data and the identication of potential predictive biomarkers.