Conflict prediction using land attributes

[ChristyLazicky]

I'm trying to build a model in ArcGIS to predict whether or not a certain land feature will have a conflict event. My data is set up as follows: I have a country divided up into ~700 polygons. Each polygon has a binary indicator to show whether it has had a conflict event (conflict=1) or not (conflict=0) in the past year. Each polygon feature also has many other attributes that I'd like to use for the prediction (e.g. land type, elevation, population, income, food security, etc.). I'd like to build a model using roughly 70% of the polygon features and then test the model's accuracy on the remaining 30%. I'm separately building a model in Python but I also wanted to use ArcGIS to take the spatial element of conflict into account.

What is the best tool in ArcGIS to do this? I've been looking at the Geostatistical Analyst toolbox but it seems like those models only do extrapolation / prediction using information from the outcome variable (and don't take other attributes about the feature into account). I also know I can use GWR to assess the impact of each predictor variable but I'm not sure if I can use that to predict conflict on a test set of data and measure the model's accuracy.

Any feedback would be greatly appreciated! Thank you!

[DanPatterson_Retired]

what are the root causes of the conflict?  I know you have a bunch of variables, but it would be best to test a model, rather than find data that may coincidently mirror the primary thesis.  ie Elevation...  how are you quantifying food security? etc

[ChristyLazicky]

I'm looking at conflict in the Democratic Republic of Congo, Central African Republic, and Somalia so the roots are incredibly complex. I should have clarified that I am looking to do this to improve conflict early warning systems so the task is less about understanding what are the root causes (because they will never be solved by any simple policy intervention) and more about building a model that is highly predictive of conflict in order to allocate peacekeeping resources better.

Do you know which tools would be appropriate in ArcGIS for this task? Thanks!

[DanPatterson_Retired]

An overview of the Spatial Statistics toolbox—Help | ArcGIS for Desktop 

would be a more appropriate location.  Perhaps looking for spatial patterns might allow you to examine whether there is any.  However, most of the statistical tests are predicated on the concept of spatial randomness... ie the phenomenon has a 'similar' chance of occuring and not occuring.  If clustering appears or is suggested, then perhaps there is something in the landscape.  Whether there is something about the landscape itself or the objects on it, is the base of the modelling.  I would caution about using such studies for predictive modelling, but that is your call.  When dealing with people, nothing is really predictable, except unpredictability itself.

The short summary, the geostatistical analyst is probably not the toolset you are interested in.  Basic statistical tests focused on attributes may be more appropriate in the first instance.  Good luck