| Popis: |
Wireless sensor networks (WSNs) have got tremendous interest from various real-life appliances, in particular environmental applications. In such long-stand employed sensors, it is difficult to check the features and quality of raw sensed data. After the deployment, there are chances that sensor nodes may expose to unsympathetic circumstances, which result in sensors to stop working or convey them to send inaccurate data. If such things not detected, the quality of the sensor network can be greatly reduced. Outlier detection ensures the quality of the sensor by safe and sound monitoring as well as consistent detection of attractive and important events. In this article, we proposed a novel method called smooth auto-encoder to learn strong plus discriminative feature representations, and reconstruction error of among input–output of smooth auto-encoder is utilized as an activation signal for outlier detection. Moreover, we employed LSTM-bidirectional RNN for maturity voting for collective outlier detection. |
