Explaining and Predicting Peacekeeping Operations Effectiveness: A Collaborative Effort between Oxford University and Zhejiang University

The Department of Politics and International Relations at Oxford in collaboration with University Zhejiang University's Center for Data Science, has pioneered groundbreaking research focused on the effectiveness of peacekeeping operations (PKOs) from 1994 to 2019. Led by researchers Andrea Ruggeri (Oxford University) and Andre Python (Zhejiang University), this study utilizes interpretable machine learning algorithms to predict and explain conflict a year ahead, within 1x1 degree grid cells in countries that experienced PKOs. This work was presented at the United Nations in New York on 16 February 2023 in an internal meeting with the Early Warning Team of the United Nations Department of Political and Peacebuilding Affairs.

The research highlights the importance of quantifying the effects of PKOs on conflict in both space and time, enabling policymakers to design and implement efficient measures for conflict resolution. By employing Gini importance and accumulated local effects plots, the research team identified main conflict predictors and quantified their effects. The findings suggest a logarithmic association between time to PKOs and conflict risk, with larger effects observed when conflict in a grid cell may spill over into neighboring cells. The study also revealed an inverse U-shape pattern when the definition of conflict was restricted to one-sided violence. These insights have significant implications for refining theories linking peacekeeping operations and conflict, as well as providing guidance for policymakers on the ground.

This project serves as an excellent example of how theoretically-informed interpretable machine learning algorithms can be used to predict and explain complex forms of political violence at policy-relevant scales. Furthermore, the interpretability of these algorithms can foster trust between modelers and policymakers, a crucial factor for realizing the full potential of these promising techniques. Zhejiang University is proud to be a part of this impactful collaboration with Oxford University, contributing to our understanding of peacekeeping operations and their effects on conflict. As we continue to explore innovative ways to address global challenges, we remain committed to working together with leading institutions and researchers to make a meaningful difference in the world.

牛津大学政治与国际关系系与浙江大学数据科学中心合作，开展了一项关于1994年至2019年间维和行动（PKOs）有效性的开创性研究。该研究由研究员Andrea Ruggeri（牛津大学）和Andre Python（浙江大学）带领，利用可解释的机器学习算法预测和解释提前一年的冲突，针对在1x1度网格单元内经历PKOs的国家。这项工作于2023年2月16日在联合国纽约总部的一个内部会议上向联合国政治和和平建设事务部的预警团队介绍。

该研究强调了在空间和时间上量化PKOs对冲突影响的重要性，使政策制定者能够设计和实施有效的冲突解决措施。通过采用基尼重要性和累积局部效应图，研究团队确定了主要冲突预测因素并量化了其效果。研究结果表明，PKOs时间与冲突风险之间存在对数关联，当网格单元内的冲突可能溢出到相邻单元时，观察到的效果更大。此外，当冲突定义仅限于单方面暴力时，研究还发现了一种反U型模式。这些见解对于完善维和行动与冲突之间的理论联系以及为一线政策制定者提供指导具有重要意义。

该项目是一个很好的例子，说明了如何使用理论上的可解释机器学习算法来预测和解释政策相关规模的复杂政治暴力形式。此外，这些算法的可解释性可以增进模型师和政策制定者之间的信任，这是实现这些有前景技术的全部潜力的关键因素。浙江大学为能与牛津大学进行这一具有影响力的合作而感到自豪，共同提高我们对维和行动及其对冲突影响的理解。在继续探索应对全球挑战的创新方法时，我们将继续与领先的机构和研究人员合作，为世界作出有意义的贡献。