Modelling militarized interstate disputes using data mining techniques: Prevention and prediction of conflicts

Abstract

The use of modern data mining techniques on large datasets has become a recent phenomenon across a broad range of applications. One of the most frequent tasks is to build statistical models using historical data and utilize them to predict new, so far unclassified, cases. This article examines the problem of predicting a military interstate dispute between two states (dyad) by employing selected data mining techniques. Suitable methods are identified and applied to the existing dataset of politically relevant dyads. The result is the building of statistical models for the classification of potential dyadic conflicts. The overall performance of these models is verified and cost analysis is done based on the different impacts of incorrect classification. The results are compared with those of other published research studies in the field of conflict prediction; the models created by data mining techniques significantly outperform all rival algorithms and approaches. Finally, the last part of the article presents the results of applying data mining techniques to association, i.e. to discovering relationships and dependencies in the data.