Improving IoT data availability via feedback- and voting-based anomaly imputation

In the era of the Internet of Everything, the massive Internet of Things (IoT) data bring multiple challenges to IoT development, high system availability, and ultimately the quality of service (QoS) improvement of IoT systems. Abnormal data is one of the main reasons that lead to unexpected processing results and fatal functionality faults in IoT systems, which also hinder the development and deployment of IoT. Existing outlier identification approaches detect anomalies in IoT without further processing these anomalies to improve the availability of datasets. In this paper, we develop a feedback- and voting-based anomaly imputation technique that improves IoT data availability by imputing anomalous sensor data. Specifically, we first propose a feedback- and voting-based feature selection method that extracts crucial features from raw datasets by using mini-batch data. We then utilize the selected features to detect anomalies, followed by an anomaly imputation technique that replaces abnormal data in raw datasets. Experimental results show that the proposed dataset anomaly imputation technique can achieve anomaly detection accuracy and availability improvement of up to 94.88% and 97.96%, respectively.