USOM: Mining and visualizing uncertain data based on self-organizing maps

Recently, mining uncertain data is gaining considerable attention due to more and more applications, such as sensor database, location database, biometric information systems, produce uncertain data. Though there exist a lot of approaches to cluster the uncertain data, few of them address mining and visualizing uncertain data. In this paper, we propose a new neural network algorithm called uncertain self-organizing map (USOM) which combines fuzzy distance function and self-organizing map to mine and visualize the uncertain data. The self-organizing map assigns the high dimensional data to the corresponding neurons and projects them on a low-dimensional grid which consists of the neurons. Each neuron is viewed as a small cluster which is a collection of the uncertain data. We merge the neurons in the low-dimensional grid to form the bigger clusters by minimal spanning tree. The experiments show that the new approaches works well in the uncertain dataset.