It has become a mainstream to use physical models to quantify expected energy savings from alternative retrofit methods and technologies. However, they are not suitable for predicting energy use of buildings when detailed and specified input parameters are unavailable. The overall purpose of the research is to support the stakeholders in taking decisions on refurbishments options when not all of physical information is available, in order to achieve the Swedish Energy Agency's measurements of near-zero energy buildings. The research will transfer big data from Swedish Energy Performance Certificates for building retrofitting. A Support Vector Machines and Fuzzy C-means clustering (SVM-FCM) integrated machine learning algorithm is used directly to extract the case-specific knowledge from EPC big data regarding building characteristics and energy saving of retrofit measures. It enables to prioritize retrofit measures and compute their expected energy savings for buildings. This proposed data driven method is an attempt of taking advantage of big data for practical building retrofit selection.