Title: A fuzzy clustering based data fusion method
Authors: Mika Sato-Ilic - University of Tsukuba (Japan) [presenting]
Abstract: A method of data fusion based on fuzzy clustering is proposed. Recently, data fusion has gained tremendous interest as a means to combine different datasets in big data analysis. How to include the common variables though the different datasets is an important issue in data fusion. This study takes a new approach which utilizes fuzzy clusters obtained from fuzzy clustering through different data sets as the common variables in the data fusion. Then, the clusters for each dataset are used as a scale to select the variables from the combination of all variables of all the datasets to explain each original dataset. Also, the selected variables can be base spans of a linear subspace to explain the original each dataset. Since the linear subspace can be obtained for each original dataset, the multiple linear subspaces can be obtained for the multiple datasets. Then using a projector which projects data in different datasets to the intersection of these multiple linear subspaces, the single common predicted data values through different data sources can be obtained. This method will reduce big data to a space of lower dimensions, and obtains a unique solution through multiple datasets. Numerical examples will be included.