In the field of wireless sensor network (WSN), anomaly detection is an important task. In this work, we have presented an anomaly detection process using ensemble random forest (ERF) in wireless sensor networks. We choose Decision Tree, Naive Bayes, and K-Nearest Neighbor as the base learners of the ensemble. We also used bootstrap sampling to construct the random forest. Here, we used python 3.7.7 with machine learning module sci-kit learn 0.23.1 to implement our learning algorithm. We evaluated our ERF algorithm using a real-world sensor dataset, namely activity recognition based on multi-sensor data fusion (AReM) dataset. We used performance metrics, namely, accuracy, sensitivity, specificity, precision, recall, f measure, and Gmean, to show that our novel ERF performs better than the base learners in isolation. We also showed the misclassification error for out-of-bag data.

在无线传感器网络（WSN）领域，异常检测是一项重要任务。在这项工作中，我们提出了一种在无线传感器网络中使用集成随机森林 (ERF) 的异常检测过程。我们选择决策树、朴素贝叶斯和 K-最近邻作为集成的基学习器。我们还使用引导抽样来构建随机森林。在这里，我们使用 python 3.7.7 和机器学习模块 sci-kit learn 0.23.1 来实现我们的学习算法。我们使用真实世界的传感器数据集评估了我们的 ERF 算法，即基于多传感器数据融合 (AReM) 数据集的活动识别。我们使用性能指标，即准确度、灵敏度、特异性、精确度、召回率、f 度量和 Gmean，来表明我们的新 ERF 比单独的基础学习器表现更好。