The traditional methods to derive building typologies make it difficult to corelate it with the energy performance. The traditional typologies are based on observable or known construction parameters, it is normally defined based on the age of dwellings and are no longer valid. The research proposes the use of machine learning and unsupervised clustering methods on the validated Energy Performance Certificate (EPC) database to derive new and relevant dwelling segmentation approach. The aim to create a real-time building stock model based on parameters within EPC database for robust energy analysis. The research will explore the potential of implementation of machine learning and data mining to improve the credibility and utility of the stock model characterized using EPC database.