In the recent years, several citizen science platforms for biodiversity monitoring have emerged. These platforms represent a powerful tool for collecting biodiversity data for researchers and increasing the knowledge of participants. Typical biodiversity data are species names observed at a given time and place by numerous participants. The use of photos to document observations allows data validation, in particular validation of species identification, a key aspect needed for the quality control of such databases. However, the increasing amount of data collected represents a major challenge given the limited number of co-opted experts dedicated to data validation. Therefore, detecting miss identifications can be very helpful to focus the limited expert workforce on dubious identifications. In this paper, we test various machine learning approaches to detect miss-identifications in such databases based on various features extracted form the history of validated observations. The proposed model can be used to automate the data validation process in the SPIPOLL platform.

近年来，出现了一些用于生物多样性监测的公民科学平台。这些平台是为研究人员收集生物多样性数据并增加参与者知识的强大工具。典型的生物多样性数据是许多参与者在给定时间和地点观察到的物种名称。使用照片记录观察结果可以进行数据验证，尤其是物种识别的验证，这是此类数据库质量控制所需的关键方面。但是，鉴于被选为数据验证专家的人数有限，收集到的数据量不断增加是一项重大挑战。因此，检测未命中的标识对于将有限的专家队伍集中在可疑标识上会很有帮助。在本文中，我们基于从经过验证的观测历史中提取的各种特征，测试了各种机器学习方法来检测此类数据库中的遗漏识别。所提出的模型可用于在SPIPOLL平台中自动化数据验证过程。