The ability to rapidly assess the spatial distribution and severity of building damage is essential to post-event emergency response and recovery. Visually identifying and classifying individual building damage requires significant time and personnel resources and can last for months after the event. This article evaluates the feasibility of using machine learning techniques such as discriminant analysis, k-nearest neighbors, decision trees, and random forests, to rapidly predict earthquake-induced building damage. Data from the 2014 South Napa earthquake are used for the study where building damage is classified based on the assigned Applied Technology Council (ATC)-20 tag (red, yellow, and green). Spectral acceleration at a period of 0.3 s, fault distance, and several building specific characteristics (e.g. age, floor area, presence of plan irregularity) are used as features or predictor variables for the machine learning models. A portion of the damage data from the Napa earthquake is used to obtain the forecast model, and the performance of each machine learning technique is evaluated using the remaining (test) data. It is noted that the random forest algorithm can accurately predict the assigned tags for 66% of the buildings in the test dataset.

中文翻译：

---

使用机器学习对建筑物的地震损坏进行分类

快速评估建筑物损坏的空间分布和严重程度的能力对于事后应急响应和恢复至关重要。目视识别和分类单个建筑物损坏需要大量的时间和人力资源，并且可能在事件发生后持续数月。本文评估了使用机器学习技术（例如判别分析、k 最近邻、决策树和随机森林）来快速预测地震引起的建筑物损坏的可行性。研究使用了 2014 年南纳帕地震的数据，根据分配的应用技术委员会 (ATC)-20 标签（红色、黄色和绿色）对建筑物损坏进行分类。0.3 s 周期内的频谱加速度、断层距离和若干建筑物特定特征（例如年龄、建筑面积、计划不规则的存在）被用作机器学习模型的特征或预测变量。纳帕地震的部分破坏数据用于获取预测模型，并使用剩余（测试）数据评估每种机器学习技术的性能。值得注意的是，随机森林算法可以准确预测测试数据集中 66% 的建筑物的分配标签。