Spatial Temporal Models

The Problem

How to Make a Complex Model Easy to Understand?

The Statistics Group at IBM developed a group of Geospatial predictive models.  These models were extremely complex models that required a multiple step 

The Process

Technical Issues

These were complex concepts that required us to understand how to get the data into the model, what the results mean, and how they were used to answer questions about the data.  We had a series of discussions and training with the statisticians to understand the requirements, parameters, and limitations of the analysis. 

From those discussions, we faced the following challenges:

• Geospatial files required significant preparation before they could be used in the analysis.

• The procedure required the data be formatted in a very specific way that required an entirely separate "Data Preparation" tool. 

• The analysis has a large number of difficult to define parameters.   

User Testing

Given the relative complexity of the analytics involved, it was a challenge to find users who were knowledgeable enough to know how to use the procedure.  We managed to find a small group of users and tested with them to validate the designs.

In the end, with the complexity we had to create a "Wizard-Like" step by step experience where each step was validated.

The Product

Make the Conclusions Easy to Read

For the output of the procedure, we created an interactive output that allowed for drilldown and highlighted the key findings of the model.  The output would identify the 'predictive rules' for the different regions on the map and display them based on their relative predictive strength.