Visual Data Mining
This project seeks to bring highly interactive visual analytics to ecological forecasting. In particular, we are collaborating with a group of ecologists and ornithologists to extract new and useful patterns out of bird migration data. We are applying clustering on terrestrial ecosystem (TECO) data as part of a visual data mining process. Our purpose in doing clustering is to discover aspects of the data which are either completely new, or which are already suspected to exist or which are hoped not to exist. For example, the users may navigate through the data and want to view the clustering output in any of the above stated ways. Our research directions include: 1. Showing the change of shape and size of the clusters with time. As the user interacts with the data in any of the above stated ways, the clusters should be identified spatially. In this way, by interacting with the range of data points, users can visualize the shape and size of clusters (we encode each cluster with different colors using a qualitative scheme from ColorBrewer). 2. Decreasing the amount of time it takes to calculate the clusters. Whenever the time range of data points changes interactively, clustering should not start over from the beginning. It should instead reuse information about prior clustering performed on time ranges that overlap with the current query. Using a caching scheme on color centroids, we hope to substantially decrease the amount of processing time that clustering algorithms take to find new clusters.