The Promise and Pitfalls of Conflict Prediction: Evidence from Colombia and Indonesia
利用哥伦比亚和印度尼西亚的精细数据，尝试用机器学习预测一年后的暴力事件，发现模型能识别持续高暴力热点，但难以预测新爆发或升级，表明年度数据预警系统尚不可行。
Abstract How feasible is violence early-warning prediction? Colombia and Indonesia have unusually fine-grained data. We assemble two decades of local violent events alongside hundreds of annual risk factors. We attempt to predict violence one year ahead with a range of machine learning techniques. Our models reliably identify persistent, high-violence hot spots. Violence is not simply autoregressive, as detailed histories of disaggregated violence perform best, but socioeconomic data substitute well for these histories. Even with unusually rich data, however, our models poorly predict new outbreaks or escalations of violence. These “best-case” scenarios with annual data fall short of workable early-warning systems.