The aim of this project was to provide a hazard score for all locations in Ukraine that were involved in war-related events such as battles, explosions/remote violence events, violence against civilians and factors that pose risk to civilians living in those locations. The idea is to take those factors and condense them into one readable and comprehensible score which would reflect the danger level of the associated location.

The main issue I have faced while developing a statistical method that could provide such a score was to combine domain knowledge and account for the known unknowns. In complex events like wars, it's very difficult to obtain a comprehensive dataset that can capture the underlying dynamics. Not because of lack of data but due to the abundance and sheer number of factors involved that are either too hard or too expensive to account for.

The issue of inherent bias involved in various sources is also prevalent and poses a serious risk of polluting a dataset. Hence, besides accounting for all factors that contribute to an event, one must also consider how valid the data that they’re using is. Therefore, I decided to use the data provided to me by ACLED for non-commercial purposes, the aim of this project is for data exploration, visualization and awareness. ACLED is a non-profit organization that has been collecting data on armed conflict events since 2014.

ACLED’s raw dataset was comprehensive enough to include three levels of administrative borders for each location, event types, sub events, dates, casualties, presence of civilian targeting, what actors were involved in the event, and a text based note column that included additional information pertaining to the event.

As mentioned before, due the complexity and nature of armed conflicts there are a significant number of factors that cannot be accounted for and some are just not available to the public. Namely for this project, ideally, I would need access to air defense numbers in a location, overall equipment concentration, the number of missiles that have been launched in a remote violence event, where they were launched from, how many have been shot down, logistic routes and hubs and this is just to name a few. Some of this data is available to the public, but the sources are sparse and inconsistent. In other words, the known unknowns. Hence, for now I have decided to use the baseline data set from ACLED and process the data enough to account for some of those factors I mentioned earlier.

Quick note on the project's implementation: everything was done using python and various data processing libraries, numpy and pandas being at the core of it and plotly’s dash was used for making the web app.