Fast Stepwise Regression on Linked Data

The main focus of research in machine learning and statistics is on building more advanced and complex models. However, in practice it is often much more important to use the right variables. One may hope that recent popularity of open data would allow researchers to easily find relevant variables. However current linked data methodology is not suitable for this purpose since the number of matching datasets is often overwhelming. This paper proposes a method using correlation based indexing of linked datasets which can significantly speed up feature selection based on classical stepwise regression procedure. The technique is efficient enough to be applied at interactive speed to huge collections of publicly available linked open data.