Event
The Triangle of Madness: Identifying the three Types of Armed Conflict from Data
- 12 July 2024
- Expired!
- 3:00 pm - 3:30 pm
Location
- Attendance: in person
- Language: EN
Event
The Triangle of Madness: Identifying the three Types of Armed Conflict from Data
The drivers of armed conflict are manifold, broadly including economic interests, ethnic strife, and political tension. Still, the sheer variety of proposed mechanisms presents a challenge for organizing them into an integrated framework facilitating the attribution of particular mechanisms to specific conflicts.
We take a step in this direction by developing an unsupervised machine learning approach to identify major classes of conflict that then lend themselves to plausible mechanistic hypotheses. We combine vast data across spatial and temporal resolutions spanning climate, geography, infrastructure, economics, demographics, and population to obtain a comprehensive picture of candidate factors that matter for conflict. We combine the diverse data sets with a multiscale clustering technique for joining together conflict events in adjacent regions based on information theoretic proxies for causality.
As a result, we discover three archetypes of conflicts: major unrest, local conflicts, and spillovers/sporadic conflicts; along with correlated conflict properties of each class, enabling us to extract mechanistic hypotheses tailored to each class. We observe that major unrest predominantly propagates in areas characterized by high population, well-developed infrastructure, and favorable geographic conditions, while demographic and climatic factors appear to play a less significant role.
Conversely, smaller conflicts are less likely to escalate into major unrest unless they impact areas with high populations irrespective of other factors. Additionally, sporadic spillovers of conflicts and tensions are typically observed in regions with low populations, inadequate infrastructure, and poor economic conditions and these conflicts don’t evolve into bigger conflicts.
In conclusion, we develop a data-oriented methodology for classifying distributed social processes, address a need for systematic categorization of armed conflict such as in humanitarian law, and unearth mechanisms based on empirical analysis.