关键词: 异常检测, 随机森林, 无线传感器网络（WSN）, 萤火虫算法, 选择性集成学习

Abstract: Data anomaly detection plays an important role in wireless sensor network (WSN) application systems for environmental monitoring. The redundant decision trees decrease the efficiency of anomaly detection algorithm based on traditional random forest (RF). To address this issue, a new anomaly detection algorithm (MBGSO-ARF) for WSNs using the theory of selective ensemble is proposed in this paper. MBGSO-ARF is based on mutation binary glowworm swarm optimization (MBGSO) and adaptive updating random forest (ARF) algorithm. MBGSO-ARF uses the improved BGSO algorithm to optimize the RF for selective ensemble to obtain the optimal sub-ensemble classifiers, which reduces the ensemble scale of the detection model, and can update the model adaptively with the evolution of data stream. Thus the accuracy of anomaly detection is improved and the time cost is reduced. An experimental comparison between MBGSO and binary particle swarm optimization (BPSO) is given. Simulation results show that the proposed algorithm is superior to BPSO algorithm, and MBGSO-ARF algorithm has higher accuracy than the other algorithms. And the size of the ensemble model is compressed. The above results demonstrate the effectiveness of MBGSO-ARF.

Key words: anomaly detection, random forest, wireless sensor network (WSN), glowworm swarm optimization, selective ensemble learning