Artificial Intelligence for Conflict Management

TL;DR

This paper compares Support Vector Machines and neural networks for predicting militarised interstate disputes, finding SVMs outperform NNs in accuracy but NNs offer more interpretability and consistency.

Contribution

It introduces the use of SVMs for MID prediction and compares their performance with neural networks, highlighting their respective advantages.

Findings

SVMs outperform NNs in MID prediction accuracy.

Neural networks provide more consistent and interpretable sensitivity analysis.

SVMs are more effective for predictive accuracy in this context.

Abstract

Militarised conflict is one of the risks that have a significant impact on society. Militarised Interstate Dispute (MID) is defined as an outcome of interstate interactions, which result on either peace or conflict. Effective prediction of the possibility of conflict between states is an important decision support tool for policy makers. In a previous research, neural networks (NNs) have been implemented to predict the MID. Support Vector Machines (SVMs) have proven to be very good prediction techniques and are introduced for the prediction of MIDs in this study and compared to neural networks. The results show that SVMs predict MID better than NNs while NNs give more consistent and easy to interpret sensitivity analysis than SVMs.