Today’s forecasting techniques, which are integrated into several information systems, often use ensembles that represent different scenarios. Aggregating these forecasts is a challenging task: when using the mean or median (common practice), important information is lost, especially if the underlying distribution at every step is multimodal. To avoid this, the authors present a heatmap visualization approach. It is easy to visually distinguish regions of high activity (high probability of realization) from regions of low activity. This form of visualization allows to identify splitting paths in the forecast ensemble and adds a “third alternative” to the decision space. Most forecast systems only offer “up” or “down”: the presented heatmap visualization additionally introduces “don’t know”. Looking at the heatmap, regions can be identified in which the underlying forecast model cannot predict the outcome. The authors present a software prototype with interactive visualization to support decision makers and discuss the information gained by its use. The prototype has already been presented to and discussed with researchers and practitioners.
Köpp, Cornelius; Mettenheim, Hans-Jörg von; and Breitner, Michael H.
"Decision Analytics with Heatmap Visualization for Multi-step Ensemble Data - An Application of UncertaintyModeling to Historical Consistent Neural Network and Other Forecasts,"
Business & Information Systems Engineering:
Vol. 6: Iss. 3, 131-140.
Available at: https://aisel.aisnet.org/bise/vol6/iss3/2