The main objective of predictive methods is the search for optimal models for various modeling techniques: classical (multiple regression, discriminant analysis), less classic (classification and regression tree) or machine learning (neural networks, ensemble methods, support vector machines). The article focuses on an uniform and synthetic presentation for the supervised learning methods the most commonly used for knowledge discovery from (very) large amount of data (Big data, KDD) for decision support in various fields of application. For each method were highlighted, as appropriate, a number of specific issues, essential for an data prospector: the fields of the application, the significances of the coefficients , the discrimination power of the characteristics, the methods for selection of variables, the appropriateness of the model with the observed data, the performances measurement, the separation of model estimation error from the prediction estimation errors, over-learning control, characterization and interpretation of results, computational performances.