Variational learning of a shifted scaled Dirichlet model with component splitting approach

Mixture models have become arguably one of the most widely used statistical approaches to perform inference on various types of data and have been successfully applied in data mining and numerous real world applications. In this work, we focus on the variational learning of finite shifted scaled Dirichlet mixture models. Component splitting is one of the major assets of our model which prevents over-fitting. Furthermore, the number of components can be estimated automatically and simultaneously along with parameters estimation. The performance and effectiveness of proposed model is verified by experimenting on two real-life applications, namely, occupancy estimation and activities recognition in smart buildings.