BN-Mapping:Visual Analysis of Geospatial Data with Bayesian Network
Wei CHEN,Biao ZHU,Hong-Xin ZHANG
DOI: https://doi.org/10.11897/SP.J.1016.2016.01281
2016-01-01
Abstract:Geospatial data is those data with explicit geographic position information.Usually,it is multi-source as well as multivariate.Analyzing multivariate geospatial data is challenging because the causal relationships among multiple variates are location-dependent,heterogeneous and multi-faceted.In this paper,we propose a novel visual analysis approach for geospatial data based on Bayesian Network (BN).Firstly,we propose a heuristic discretion method,which discretizes the original continuous geospatial data into a discrete one;Then,a Bayesian Network model is learned from the discretized multivariate geospatial dataset;After that,we designed three visual representations,namely,a global causal relationship in the form of a graph,detailed relationship quantities and probability in the form of a pixel-map,and a salient BN-distribution in the form of pixel charts,for the topology and conditional probability table of the learned Bayesian Network, and the geospatial distribution,respectively.Finally,we designed an interactive visual analysis interface,which well combines all the views and allows users to interactively explore,discover and reason about implicit patterns in a location-aware fashion.Case studies on several datasets for various analysis tasks,including anomaly detection,location-aware classification,and causal relationship discovery,demonstrate the efficiency of our approach in analyzing geospatial data.