A computer-implemented method, system and computer program product for providing predictions tailored for a specific domain. Each field within a set of datasets is classified with a concept defined in a domain specific ontology. After receiving a target field and identifying a target concept tagged to the target field, concepts that have an influential impact directly to the target concept or indirectly to a concept tagged to a field associated with the target field are identified. Fields are then identified as candidates for predictors from the identified concepts using a semantic relationship and a numeric association combined approach. An influential confidence score and a numeric analytics score are then calculated for each of these candidates upon which a final confidence score is computed. The identified candidates are ranked based on the final confidence scores, and the top ranked candidates for predictors are selected to build a statistical model.