Consisting of over 1,400 species, bats are the second most diverse group of mammals. Many species are currently threatened, while another ~244 species are currently listed as Data Deficient by International Union for Conservation Science (IUCN) Global Red List. IUCN assessments can be vital for early conservation intervention and would be aided by a rapid preliminary tool to determine priority for further research and full IUCN Red List assessments. While some tools currently exist to generate extinction risk predictions for bats, they generally require difficult to obtain data, such as phylogenies and trait data. New approaches exist that utilize machine learning algorithms, such as random forest and neural networks, and can accomplish the same task using easier to obtain occurrence-derived data. Here, I fit models that can predict a species’ potential IUCN Red List category using prior assignments (critically endangered, endangered, vulnerable, near threatened, and least concern) grouped into binary categories of “Not Threatened” and “Threatened” as training data and applied the best approach to some Data Deficient bat species. These classifications can be used to prioritize investments in conservation for these species. Methods used included index-based approaches (ConR and rCat) and machine learning (IUC-NN and random forest). The best performing model used a random forest algorithm and could accurately predict IUCN binary categories (“Threatened” and “Not Threatened”) 86.9% of the time. While the overall accuracy is similar to the other approaches used here, it vastly outperforms when looking at other metrics like false-negative rate (incorrectly listing a “Threatened” species as “Not threatened”). As a result, this approach could be used as a first step to predict possible IUCN categories for bats that could be used to prioritize conservation research and is not a replacement for full IUCN Global Red List assessment into extinction risk categories.

中文翻译：

---

比较来自机器学习和其他蝙蝠方法的 IUCN 红色名录类别的预测

蝙蝠由 1,400 多种物种组成，是第二大种类繁多的哺乳动物。许多物种目前受到威胁，而另外约 244 个物种目前被国际保护科学联盟 (IUCN) 全球红色名录列为数据不足。IUCN 评估对于早期保护干预至关重要，并且将得到一个快速初步工具的帮助，以确定进一步研究和全面 IUCN 红色名录评估的优先级。虽然目前存在一些工具来生成蝙蝠灭绝风险预测，但它们通常需要难以获得数据，例如系统发育和性状数据。存在利用机器学习算法的新方法，例如随机森林和神经网络，并且可以使用更容易获得事件衍生数据来完成相同的任务。这里，我拟合的模型可以预测一个物种的潜在 IUCN 红色名录类别，使用之前的分配（极度濒危、濒危、易受威胁、近危和最不关心）分组为“未受威胁”和“受威胁”的二元类别作为训练数据并应用一些数据缺乏蝙蝠物种的最佳方法。这些分类可用于优先投资保护这些物种。使用的方法包括基于索引的方法（ConR 和 rCat）和机器学习（IUC-NN 和随机森林）。性能最佳的模型使用随机森林算法，可以准确预测 IUCN 二元类别（“受威胁”和“未受威胁”）的准确率达 86.9%。虽然整体准确性与此处使用的其他方法相似，在查看假阴性率（错误地将“受威胁”物种列为“未受威胁”）等其他指标时，它的表现大大优于其他指标。因此，这种方法可以用作预测 IUCN 可能的蝙蝠类别的第一步，这些类别可用于优先保护研究，并且不能替代 IUCN 全球红色名录对灭绝风险类别的完整评估。