We use GPS tracking data and a machine learning approach to examine the movement of artisanal fishers from the Commonwealth of Dominica during their foraging trips. The model examines track segments previously identified as resource patches by a cumulative sum method. We build on this work by using domain knowledge to train a classification and regression tree analysis to discern different patch types, including patches associated with fish aggregating devices (FADs). We train the method by first labeling segments according to patch type using the domain knowledge ground truth data collected during participant observations. Next, we use 10 derived variables to describe the different patch types according to segment location, size, and shape as well as the speed, direction, duration, and sinuosity of the fishers’ movement while at the patches. Using these data, the classification tree creates a program that classifies the patches by type. Model testing shows that we can expect to correctly discern FAD patches for >90% of the cases for which there are no observational data. In the area of behavioral analysis, these methods can reduce costs and save time via automatic processing of increasingly big data sets to answer anthropological questions that are otherwise difficult to answer.

中文翻译：

---

使用来自多米尼克联邦手工渔民的移动数据识别补丁类型

我们使用 GPS 跟踪数据和机器学习方法来检查多米尼加联邦的手工渔民在觅食之旅中的活动。该模型通过累积总和方法检查先前标识为资源补丁的轨道段。我们通过使用领域知识来训练分类和回归树分析以识别不同的补丁类型，包括与鱼聚集设备 (FAD) 相关的补丁，以此为基础。我们首先使用参与者观察期间收集的领域知识地面实况数据根据补丁类型标记片段来训练该方法。接下来，我们使用 10 个派生变量来根据分段位置、大小和形状以及渔民在斑块上移动的速度、方向、持续时间和曲折度来描述不同的斑块类型。使用这些数据，分类树创建一个程序，按类型对补丁进行分类。模型测试表明，对于 90% 以上的没有观察数据的案例，我们可以期望正确识别 FAD 补丁。在行为分析领域，这些方法可以通过自动处理越来越大的数据集来回答难以回答的人类学问题，从而降低成本并节省时间。