Predicting Food Crises 2020, Dataset for reproducing working paper results - Afghanistan, Burkina Faso, Chad...and 18 more

Abstract
---------------------------

Globally, more than 130 million people are estimated to be in food crisis. These humanitarian disasters are associated with severe impacts on livelihoods that can reverse years of development gains. The existing outlooks of crisis-affected populations rely on expert assessment of evidence and are limited in their temporal frequency and ability to look beyond several months. This paper presents a statistical foresting approach to predict the outbreak of food crises with sufficient lead time for preventive action. Different use cases are explored related to possible alternative targeting policies and the levels at which finance is typically
unlocked. The results indicate that, particularly at longer forecasting horizons, the statistical predictions compare favorably to expert-based outlooks. The paper concludes that statistical models demonstrate good ability to detect future outbreaks of food crises and that using statistical forecasting approaches may help increase lead time for action.

摘要
---------------------------
全球范围内，估计有超过一亿三千万人处于粮食危机之中。此类人道主义灾难对生计的严重影响可能逆转数年的发展成果。现有针对受危机影响的群体的预测依赖于专家对证据的评估，其在时间频率上有所局限，且难以展望数月之外的情况。本文提出了一种统计森林方法，旨在预测粮食危机的爆发，并为预防行动提供充足的预警时间。本文探讨了与可能的替代目标政策以及通常解冻资金的层次相关的不同用例。结果表明，特别是在较长的预测范围内，统计预测与基于专家的展望相比，具有明显的优势。文章得出结论，统计模型在检测未来粮食危机爆发方面表现出良好的能力，且采用统计预测方法可能有助于增加行动的预警时间。