TY - JOUR
T1 - Forecasting hotspots - A predictive analytics approach
AU - MacIejewski, Ross
AU - Hafen, Ryan
AU - Rudolph, Stephen
AU - Larew, Stephen G.
AU - Mitchell, Michael A.
AU - Cleveland, William S.
AU - Ebert, David S.
N1 - Funding Information:
The authors would like to thank the Indiana State Department of Health and the Regenstrief Institute for their input on system development and user feedback, particularly Dr. Shaun Grannis for his feedback on this project. The Cancer Care Engineering Project is supported by the Department of Defense, Congressionally Directed Medical Research Program, Fort Detrick, MD (W81-XWH-08-1-0065) and the Regenstrief Cancer Foundation administered jointly through the Oncological Sciences Center at Purdue University and the Indiana University Simon Cancer Center. This work has been funded by the US Department of
PY - 2011
Y1 - 2011
N2 - Current visual analytics systems provide users with the means to explore trends in their data. Linked views and interactive displays provide insight into correlations among people, events, and places in space and time. Analysts search for events of interest through statistical tools linked to visual displays, drill down into the data, and form hypotheses based upon the available information. However, current systems stop short of predicting events. In spatiotemporal data, analysts are searching for regions of space and time with unusually high incidences of events (hotspots). In the cases where hotspots are found, analysts would like to predict how these regions may grow in order to plan resource allocation and preventative measures. Furthermore, analysts would also like to predict where future hotspots may occur. To facilitate such forecasting, we have created a predictive visual analytics toolkit that provides analysts with linked spatiotemporal and statistical analytic views. Our system models spatiotemporal events through the combination of kernel density estimation for event distribution and seasonal trend decomposition by loess smoothing for temporal predictions. We provide analysts with estimates of error in our modeling, along with spatial and temporal alerts to indicate the occurrence of statistically significant hotspots. Spatial data are distributed based on a modeling of previous event locations, thereby maintaining a temporal coherence with past events. Such tools allow analysts to perform real-time hypothesis testing, plan intervention strategies, and allocate resources to correspond to perceived threats.
AB - Current visual analytics systems provide users with the means to explore trends in their data. Linked views and interactive displays provide insight into correlations among people, events, and places in space and time. Analysts search for events of interest through statistical tools linked to visual displays, drill down into the data, and form hypotheses based upon the available information. However, current systems stop short of predicting events. In spatiotemporal data, analysts are searching for regions of space and time with unusually high incidences of events (hotspots). In the cases where hotspots are found, analysts would like to predict how these regions may grow in order to plan resource allocation and preventative measures. Furthermore, analysts would also like to predict where future hotspots may occur. To facilitate such forecasting, we have created a predictive visual analytics toolkit that provides analysts with linked spatiotemporal and statistical analytic views. Our system models spatiotemporal events through the combination of kernel density estimation for event distribution and seasonal trend decomposition by loess smoothing for temporal predictions. We provide analysts with estimates of error in our modeling, along with spatial and temporal alerts to indicate the occurrence of statistically significant hotspots. Spatial data are distributed based on a modeling of previous event locations, thereby maintaining a temporal coherence with past events. Such tools allow analysts to perform real-time hypothesis testing, plan intervention strategies, and allocate resources to correspond to perceived threats.
KW - Predictive analytics
KW - syndromic surveillance
KW - visual analytics
UR - http://www.scopus.com/inward/record.url?scp=79951944678&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=79951944678&partnerID=8YFLogxK
U2 - 10.1109/TVCG.2010.82
DO - 10.1109/TVCG.2010.82
M3 - Article
AN - SCOPUS:79951944678
SN - 1077-2626
VL - 17
SP - 440
EP - 453
JO - IEEE Transactions on Visualization and Computer Graphics
JF - IEEE Transactions on Visualization and Computer Graphics
IS - 4
M1 - 5473230
ER -