A decade’s worth of conflict data is forcing researchers to confront uncomfortable truths. That is, the war we think we understand can completely defy predictions.
Eddie Lee from Vienna’s Complexity Science Hub explains a problem that has puzzled conflict researchers for decades. “Experts have incredible intuition about conflicts and regions they know well,” he says. But there’s a catch. Every year, tens of thousands of conflicts occur around the world. “No single expert can understand all of this complexity.”
This gap between human judgment and sheer volume is where Niraj Kushwaha and his colleagues found their way. Instead of asking experts to sort conflicts into categories, why not let the data do it for you? They collected nearly 30 years of conflict records from across Africa, added information on climate, economy, geography, population density, infrastructure, and demographics, and deployed machine learning to see what they uncovered.
What I got back was surprisingly simple. Three types of conflict appear again and again across different scales of analysis. Large-scale unrest is spreading across densely populated and well-connected areas. Think of Boko Haram in Nigeria or the civil war in the Central African Republic. These persist for years and spread across borders. Local conflicts remain within a single country and typically last months rather than years, such as clashes between pro-Seleka and anti-Balaka forces or inter-clan violence in certain parts of Somalia. Additionally, there are sporadic explosions and short-term flare-ups in remote and undeveloped areas, such as the spillover of al-Shabab activities to certain parts of the Horn of Africa.
“Our algorithmic method learns what competition types should be by letting the data speak for itself,” explains Lee. “And the result is surprisingly simple.”
The team documented consistent patterns in how these three archetypes cluster across geographic and economic space. Population density and infrastructure development most clearly distinguish the major insecurity categories. Economics and geography are further subdivided. “These three conflict types naturally emerged from our data over and over again, even when we changed the analysis and the spatial and temporal scales of the data coverage,” Kushwaha said.
This is the part where everything needs to fall into place. By better classifying conflicts, you can better predict their severity. Knowing the type of violence you are facing should help you predict how many people will die, how long it will last, and what area it will cover.
It should. But that’s not the case.
When Kushwaha and his team examined whether their classification could predict the intensity of conflict (number of deaths, duration, geographic spread), they found little correlation. Knowing which archetype a conflict belongs to tells us little about how destructive it will be. “This seems counterintuitive,” Kushwaha said. “You might think that better classification would help with predictions, but the data shows that these are fundamentally different problems.”
Lee frames this finding more precisely. “Many widely used indicators and datasets may not actually improve our ability to predict how intense a conflict will escalate, suggesting the need for new approaches rather than more of the same data.”
This is not a failure of their algorithm. The research team validated their results using a random forest classifier (an entirely different machine learning approach) and got the same results. The pattern is there. Prototypes are realistic and reproducible. it’s just a matter of understanding it Kindness The conflict you are in gives you little power to predict what will happen. damage It makes an impact.
Its influence quietly spreads outward. Conflict researchers have long assumed that the geographic, economic, and demographic factors that shape conflict types also shape conflict severity. It was thought that if sufficient resources were devoted to better classification systems, early warning would be improved. you will save a life. Data suggests otherwise. Understanding the underlying conditions that give rise to different types of conflict is of no use in predicting their destructive intensity.
Woy Sok Oh, one of the team members from the University of Waterloo, highlights what this means in practice. “Different types of violence occur in different contexts, and each requires a different response.” Large-scale unrest in dense urban areas requires fundamentally different humanitarian and policy approaches than localized clan conflicts in developing regions. Classification itself has value.
But can we predict the impact? It remains elusive. “It is important to recognize that many widely used indicators may not actually improve our ability to predict how intense a conflict will become,” warns Lee. The researchers are not saying that predictive models are without value. Random forests still beat random guessing. They say more disturbing things. A feature that competitive analysts have focused on for decades may be optimized for the wrong results.
Kushwaha sees this as an opportunity rather than a dead end. “We’re investigating the limits of what can be predicted, and in doing so we hope to provide a foundation for future research.” The study revealed as much about the data that doesn’t exist as it does about the data that does exist. What is missing from global datasets? What information can actually improve conflict prediction? These are the real questions.
On maps, the three conflict archetypes often overlap within the same region. Around Mogadishu, large-scale insurgency by al-Shabaab occurs in close proximity to smaller local conflicts and sporadic incidents. The tri-border region bordered by Burundi, Rwanda, and the Democratic Republic of the Congo is home to both large-scale cross-border violence and small-scale conflicts confined within the same geography. The coexistence of different types of violence arising from similar situations adds a further layer to the mystery.
The researchers document and publish their methods and datasets, and encourage other researchers to build on their work. Their approach integrates multiple types of granular data and has the potential to reshape the way competitive analysis is conducted. But first, the field may need to readjust its expectations about what can be achieved with better data alone.
“We are not simply categorizing conflicts,” Kushwaha concluded. Perhaps the deeper lesson is that violence, even when categorized into neat archetypes, resists the quantitative predictions we have long assumed are possible given sufficient information. The data speaks clearly to this point.
study: “Data-driven conflict classification reveals weak predictive indicators”, published in Royal Society Open Science, DOI: 10.1098/rsos.250897
If our reporting has informed or inspired you, please consider making a donation. Every contribution, however big or small, helps us continue to provide accurate, engaging and trustworthy science and medical news. Independent journalism takes time, effort and resources. Your support allows us to continue uncovering the stories that matter most to you.
Join us in making knowledge accessible and impactful. Thank you for your cooperation!
